DELE ST1504 CA1 Part A: Convolutional Neural Network

---

Name: Lee Hong Yi
Admin No: 2223010
Class: DAAA/FT/2B/07

---

Objective:
To build image classifiers using deep learning networks for 15 types of vegetables. Two different input sizes (31x31 pixels and 128x128 pixels) will be considered, and the images will be converted to grayscale (1 channel) for both input sizes. The classification accuracies of the two models will be compared and discussed.

Background Information:
In this task, we will focus on developing image classifiers for a dataset containing 15 types of vegetables. The goal is to accurately classify the vegetables based on grayscale images with two different input sizes: 31x31 pixels and 128x128 pixels. By comparing the performance of models trained with these different input sizes, we can gain insights into the impact of image resolution on classification accuracy. This assignment's constraints are done to represent any difficuties that are possible in real-life data gathering, such as in the medical field.

Additional Information:
-nil-

Output Variable:
The main output variable of this task is the classification accuracy of the two models for each input size (31x31 and 128x128 pixels). Additionally, the task aims to provide insights and a discussion on the performance differences between the two models and their suitability for the given image classification problem.

---

---

Setting up required libraries.

---

import gc
import numpy as np
import pandas as pd
import seaborn as sns
import tensorflow as tf
import matplotlib.pyplot as plt
from collections import Counter
from itertools import permutations 
import os, random, pathlib, warnings, itertools, math
import CNN_function as cnnf


#importing tensorflow libraries 
from tensorflow.keras.layers import Input
from tensorflow.keras.layers import Layer
from tensorflow.keras.layers import Dense
from tensorflow.keras.layers import Conv2D
from tensorflow.keras.layers import Dropout
from tensorflow.keras.layers import Flatten
from tensorflow.keras.layers import ReLU
from tensorflow.keras.utils import plot_model
from tensorflow.keras.models import load_model
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import concatenate
from tensorflow.keras.applications import VGG16
from tensorflow.keras.models import clone_model
from tensorflow.keras.layers import MaxPooling2D
from tensorflow.keras.layers import AveragePooling2D
from tensorflow.keras.callbacks import EarlyStopping
from tensorflow.keras.layers import BatchNormalization
from tensorflow.keras.layers import GlobalAveragePooling2D
from tensorflow.keras.layers import ActivityRegularization
from tensorflow.keras.preprocessing.image import ImageDataGenerator

#importing other libraries
from IPython.display import clear_output
from sklearn.metrics import confusion_matrix
from sklearn.preprocessing import MultiLabelBinarizer
from sklearn.utils.class_weight import compute_class_weight


#ignoring redundant warnings
from warnings import simplefilter
simplefilter(action='ignore', category=UserWarning)     
simplefilter(action='ignore', category=FutureWarning) 

# Fix random seed for reproducibility
seed = 1
np.random.seed(seed)
tf.random.set_seed(seed)
tf.keras.utils.set_random_seed(0)  

gpus = tf.config.experimental.list_physical_devices('GPU')
for gpu in gpus:
    print(tf.config.experimental.get_device_details(gpu))
    tf.config.experimental.set_memory_growth(gpu, True)
print(f"There are {len(gpus)} GPU(s) present.")

{'device_name': 'NVIDIA GeForce RTX 3060', 'compute_capability': (8, 6)}
There are 1 GPU(s) present.

---

Loading Data

---

dataset='./Dataset for CA1 part A/Dataset for CA1 part A'

train_folder = os.path.join(dataset,"train")
test_folder = os.path.join(dataset,"validation")
validation_folder = os.path.join(dataset,"test")

test_sample_count = pd.DataFrame(columns = ["name", "count"])
train_sample_count = pd.DataFrame(columns = ["name", "count"])
validation_sample_count = pd.DataFrame(columns = ["name", "count"])

def count_files(rootdir):
    count_dataframe = pd.DataFrame(columns=["name", "count"])
    for i, path in enumerate(pathlib.Path(rootdir).iterdir()):
        if path.is_dir():  # count files only in directories
            count = len([name for name in os.listdir(path) if os.path.isfile(os.path.join(path, name))])
            count_dataframe.loc[i] = [path.name, count]  # Using loc to add rows
    return count_dataframe

train_sample_count = count_files(train_folder)
test_sample_count = count_files(test_folder)
validation_sample_count = count_files(validation_folder)

batch_size = None

normal_train_data = tf.keras.utils.image_dataset_from_directory(train_folder, batch_size = batch_size, seed = seed, image_size = (224, 224))
normal_test_data = tf.keras.utils.image_dataset_from_directory(test_folder, batch_size = batch_size, seed = seed, image_size = (224, 224))
normal_validation_data = tf.keras.utils.image_dataset_from_directory(validation_folder, batch_size = batch_size, seed = seed, image_size = (224, 224))
normal_full_data = normal_train_data.concatenate(normal_test_data)
normal_full_data = normal_full_data.concatenate(normal_validation_data)

print(f"There are {len(normal_train_data.class_names)} classes, {list(normal_train_data.class_names)}")
class_names = list(normal_train_data.class_names)

def normalize_image(image, label):
    image = tf.cast(image, tf.float32) / 255.0
    return image, label

normal_train_data = normal_train_data.map(normalize_image)
normal_test_data = normal_test_data.map(normalize_image)
normal_validation_data = normal_validation_data.map(normalize_image)
normal_full_data = normal_full_data.map(normalize_image)

batch_size = 32
normal_test_data = normal_test_data.batch(batch_size)
normal_train_data = normal_train_data.batch(batch_size)
normal_validation_data = normal_validation_data.batch(batch_size)

Found 9028 files belonging to 15 classes.
Found 3000 files belonging to 15 classes.
Found 3000 files belonging to 15 classes.
There are 15 classes, ['Bean', 'Bitter_Gourd', 'Bottle_Gourd', 'Brinjal', 'Broccoli', 'Cabbage', 'Capsicum', 'Carrot', 'Cauliflower', 'Cucumber', 'Papaya', 'Potato', 'Pumpkin', 'Radish', 'Tomato']

---

Exploratory data analysis.

---

Creating functions to display dataset images.

Using previously created functions for EDA.

cnnf.class_image(normal_full_data, class_names, cmap = "gray")

cnnf.average_image(normal_full_data)

From the above, which is a average of all the pictures in our dataset, we can see that there are no outstanding features in the dataset. The dataset also appears to be uniform, with the dominant color being green. The dataset also seems to be homogenous, with most of the details being shared accross the images.

These may all pose challenges to us later, as the model may not have enough features to identify the different types of vegetables.

cnnf.average_images_per_class(normal_full_data, class_names)

From the above, we can see that there are 3 distinct colors in the dataset, with Capsicums, Carrots, Potatoes, and Tomatoes having high red values, while Cauliflowers, Pumpkins, and Radishes being more white, meaning that they are higher in all color channels. The rest of the vegetables have greener average images, showing that they have a higher load in the green color channel.

This means that when we greyscale images later, there may be important data lost which affects the accuracy of the model.

We can also see that most images look similar when averaged, with most just being a circular blob. This means that the shape of most vegetables also matter to the model, and might be a important component in classification.

This matters, as it means that when the images are downsized later, we may lose some data, causing the accuracy to further suffer.

fig, axs = plt.subplots(1,3, sharey=True, figsize = (20,5))

cnnf.image_count_graph(train_sample_count , ax = axs[0], set_name = "Training")
cnnf.image_count_graph(test_sample_count, ax = axs[1], set_name = "Testing")
cnnf.image_count_graph(validation_sample_count, ax = axs[2], set_name = "Validation")

plt.show()

From the above, we can see that the dataset is imbalanced.

mean_val, std_val = cnnf.compute_running_mean_std(normal_full_data)
print(f"The mean brightness of all images is {mean_val}")
print(f"The standard deviation of all images is {std_val}")

The mean brightness of all images is 0.4197029846784955
The standard deviation of all images is 0.24243864096348536

(Note: we scaled the scalars of the image to be from 0 to 1 earlier.)

From the above, we can see that the mean brightness of all images is 0.419. This tells us that the data is not extremely bright. In addition, there is also limited variation in the data given (0.23995).

---

Establishing baseline models.

---

To help us identify the best data preparation pipeline, we shall establish a baseline model, then test it out on the different steps, to see which steps are the right ones to take to improve the accuracy of the model. As a baseline model, we shall use a simple convolutional neural network, using tensorflow's Conv2D layers. These are better suited for image classification tasks compared to other layers such as the Dense layers, due to their ability to work on local regions of the image, allowing it to detect features such as edges and patterns. This allows it to understand the structure and content of an image better than other types of models. In addition, we shall use Sparse Categorical Crossentropy over other types of entropy loss functions, as it is better suited to the nature of our dataset. Sparse Categorical Crossentropy is useful when handling a larger amount of classes, like our use case here. If we were to one-hot encode the labels, like is needed for Categorical Crossentropy, we would have generated larger, sparser matrices, which are inefficent when it comes to memory usage. Sparse Categorical Crossentropy also directly uses class labels instead of one-hot encoded vectores, allowing us to skip a step in the data preparation pipeline, simplifying the data pre-preprocessing.

The formula for Sparse Categorical Crossentropy is as follows:

\begin{aligned} L &= -\sum_{i=1}^{N} \log(p_{y_i}) \\ \text{where} \\ L &\text{ is the loss} \\ N &\text{ is the number of samples in the dataset} \\ y_i &\text{ is the true class label for the } i\text{-th sample} \\ p_{y_i} &\text{ is the predicted probability of the true class } y_i \text{ for the } i\text{-th sample} \end{aligned}

Furthermore, we shall use the Adam optimizer first. ADAptive Moment Estimation is a extension of Stochastic Gradient Descent. It combines ideas from two other optimization algorithms, namely Momentum and RMSProp. It is a good choice for us to start with, given it's "self-learning" properties, which means that it's learning rates are adaptive, relative to how frequently a parameter gets updated while training, making it suitable for problems with sparse gradients or with noisy data, like ours. It is also omputationally efficient with relatively low memory requirements, making it suitable for problems with large datasets or parameters.

earlyStopper = EarlyStopping(monitor = 'val_loss', patience = 10)

num_classes = len(class_names)

baseline_model = Sequential()
baseline_model.add(Conv2D(16, (3, 3), activation='relu'))
baseline_model.add(MaxPooling2D((2, 2)))
baseline_model.add(Conv2D(32, (3, 3), activation='relu'))
baseline_model.add(GlobalAveragePooling2D())
baseline_model.add(Dense(num_classes, activation='softmax'))
baseline_model.compile(loss = "sparse_categorical_crossentropy", optimizer = "adam", metrics = ["accuracy"])

baseline31_model = clone_model(baseline_model)
baseline31luminosity_model = clone_model(baseline_model)
baseline31desaturation_model = clone_model(baseline_model)
baseline128_model = clone_model(baseline_model)
baseline128luminosity_model = clone_model(baseline_model)
baseline128desaturation_model = clone_model(baseline_model)
baseline31_model.compile(loss = "sparse_categorical_crossentropy", optimizer = "adam", metrics = ["accuracy"])
baseline31luminosity_model.compile(loss = "sparse_categorical_crossentropy", optimizer = "adam", metrics = ["accuracy"])
baseline31desaturation_model.compile(loss = "sparse_categorical_crossentropy", optimizer = "adam", metrics = ["accuracy"])
baseline128_model.compile(loss = "sparse_categorical_crossentropy", optimizer = "adam", metrics = ["accuracy"])
baseline128luminosity_model.compile(loss = "sparse_categorical_crossentropy", optimizer = "adam", metrics = ["accuracy"])
baseline128desaturation_model.compile(loss = "sparse_categorical_crossentropy", optimizer = "adam", metrics = ["accuracy"])

---

Data Preparation.

---

To prepare the data to make it ready for the model, we need to

1. Greyscale the data
2. Resize to both 31 x 31 and 128 x 128 pixels.

We shall start by resizing the data, then moving onto investigating the different methods that can be used to greyscale the data.

def resize_image(image, label, target_height=31, target_width=31):
    image = tf.image.resize(image, [target_height, target_width])
    return image, label

res31_train_data = normal_train_data.map(lambda image, label: resize_image(image, label, 31, 31), num_parallel_calls=tf.data.AUTOTUNE)
res31_test_data = normal_test_data.map(lambda image, label: resize_image(image, label, 31, 31), num_parallel_calls=tf.data.AUTOTUNE)
res31_validation_data = normal_validation_data.map(lambda image, label: resize_image(image, label, 31, 31), num_parallel_calls=tf.data.AUTOTUNE)  
res31_full_data = res31_train_data.concatenate(res31_test_data)
res31_full_data = res31_full_data.concatenate(res31_validation_data)

res128_train_data = normal_train_data.map(lambda image, label: resize_image(image, label, 128, 128), num_parallel_calls=tf.data.AUTOTUNE)
res128_test_data = normal_test_data.map(lambda image, label: resize_image(image, label, 128, 128), num_parallel_calls=tf.data.AUTOTUNE)
res128_validation_data = normal_validation_data.map(lambda image, label: resize_image(image, label, 128, 128), num_parallel_calls=tf.data.AUTOTUNE)  
res128_full_data = res128_train_data.concatenate(res128_test_data)
res128_full_data = res128_full_data.concatenate(res128_validation_data)

We shall start by using the simplest method to greyscale the images, by averaging out all 3 color channels, to obtain the average brightness of each image.

def convert_to_greyscale(image, label):
    image = tf.image.rgb_to_grayscale(image)
    return image, label

res31greyscale_train_data = res31_train_data.map(lambda image, label: convert_to_greyscale(image, label), num_parallel_calls= tf.data.AUTOTUNE)
res31greyscale_test_data = res31_test_data.map(lambda image, label: convert_to_greyscale(image, label), num_parallel_calls= tf.data.AUTOTUNE)
res31greyscale_validation_data = res31_validation_data.map(lambda image, label: convert_to_greyscale(image, label), num_parallel_calls= tf.data.AUTOTUNE)
res31greyscale_full_data = res31greyscale_train_data.concatenate(res31greyscale_test_data)
res31greyscale_full_data = res31greyscale_full_data.concatenate(res31greyscale_validation_data)
res31greyscale_full_data = res31greyscale_full_data.unbatch()


res128greyscale_train_data = res128_train_data.map(lambda image, label: convert_to_greyscale(image, label), num_parallel_calls= tf.data.AUTOTUNE)
res128greyscale_test_data = res128_test_data.map(lambda image, label: convert_to_greyscale(image, label), num_parallel_calls= tf.data.AUTOTUNE)
res128greyscale_validation_data = res128_validation_data.map(lambda image, label: convert_to_greyscale(image, label), num_parallel_calls= tf.data.AUTOTUNE)
res128greyscale_full_data = res128greyscale_train_data.concatenate(res128greyscale_test_data)
res128greyscale_full_data = res128greyscale_full_data.concatenate(res128greyscale_validation_data)
res128greyscale_full_data = res128greyscale_full_data.unbatch()

Next, we shall use the luminosity method to greyscale the images. This works by adjusting the strength of the color channels based on the perception of the human eye, since it does not respond equally to all colors. The formula for the luminosity method is as follows $$ \text Grey = 0.21×Red+0.72×Green+0.07×Blue $$

def convert_to_greyscale_luminosity(image, label):
    luminosity_weights = tf.constant([0.21, 0.72, 0.07], shape=(1, 1, 3), dtype=image.dtype)
    grayscale_image = tf.reduce_sum(image * luminosity_weights, axis=-1, keepdims=True)
    return grayscale_image, label

res31luminosity_train_data = res31_train_data.map(lambda image, label: convert_to_greyscale_luminosity(image, label), num_parallel_calls= tf.data.AUTOTUNE)
res31luminosity_test_data = res31_test_data.map(lambda image, label: convert_to_greyscale_luminosity(image, label), num_parallel_calls= tf.data.AUTOTUNE)
res31luminosity_validation_data = res31_validation_data.map(lambda image, label: convert_to_greyscale_luminosity(image, label), num_parallel_calls= tf.data.AUTOTUNE)
res31luminosity_full_data = res31luminosity_train_data.concatenate(res31luminosity_test_data)
res31luminosity_full_data = res31luminosity_full_data.concatenate(res31luminosity_validation_data)
res31luminosity_full_data = res31luminosity_full_data.unbatch()


res128luminosity_train_data = res128_train_data.map(lambda image, label: convert_to_greyscale_luminosity(image, label), num_parallel_calls= tf.data.AUTOTUNE)
res128luminosity_test_data = res128_test_data.map(lambda image, label: convert_to_greyscale_luminosity(image, label), num_parallel_calls= tf.data.AUTOTUNE)
res128luminosity_validation_data = res128_validation_data.map(lambda image, label: convert_to_greyscale_luminosity(image, label), num_parallel_calls= tf.data.AUTOTUNE)
res128luminosity_full_data = res128luminosity_train_data.concatenate(res128luminosity_test_data)
res128luminosity_full_data = res128luminosity_full_data.concatenate(res128luminosity_validation_data)
res128luminosity_full_data = res128luminosity_full_data.unbatch()

Lastly, we shall use the desaturation method to greyscale the images. The method calculates the grey level by averaging the maximum and minimum values of the RGB components for each pixel. The formula for the desaturation method is as follows: $$ \text{Gray} = \frac{\text{max(Red, Green, Blue)} + \text{min(Red, Green, Blue)}}{2} $$

def convert_to_greyscale_desaturation(image, label):
    max_rgb = tf.reduce_max(image, axis=-1, keepdims=True)
    min_rgb = tf.reduce_min(image, axis=-1, keepdims=True)
    grayscale_image = (max_rgb + min_rgb) / 2
    return grayscale_image, label

res31desaturation_train_data = res31_train_data.map(lambda image, label: convert_to_greyscale_desaturation(image, label), num_parallel_calls= tf.data.AUTOTUNE)
res31desaturation_test_data = res31_test_data.map(lambda image, label: convert_to_greyscale_desaturation(image, label), num_parallel_calls= tf.data.AUTOTUNE)
res31desaturation_validation_data = res31_validation_data.map(lambda image, label: convert_to_greyscale_desaturation(image, label), num_parallel_calls= tf.data.AUTOTUNE)
res31desaturation_full_data = res31desaturation_train_data.concatenate(res31desaturation_test_data)
res31desaturation_full_data = res31desaturation_full_data.concatenate(res31desaturation_validation_data)
res31desaturation_full_data = res31desaturation_full_data.unbatch()

res128desaturation_train_data = res128_train_data.map(lambda image, label: convert_to_greyscale_desaturation(image, label), num_parallel_calls= tf.data.AUTOTUNE)
res128desaturation_test_data = res128_test_data.map(lambda image, label: convert_to_greyscale_desaturation(image, label), num_parallel_calls= tf.data.AUTOTUNE)
res128desaturation_validation_data = res128_validation_data.map(lambda image, label: convert_to_greyscale_desaturation(image, label), num_parallel_calls= tf.data.AUTOTUNE)
res128desaturation_full_data = res128desaturation_train_data.concatenate(res128desaturation_test_data)
res128desaturation_full_data = res128desaturation_full_data.concatenate(res128desaturation_validation_data)
res128desaturation_full_data = res128desaturation_full_data.unbatch()

---

Greyscale validation (31 x 31 image)

---

history = baseline31_model.fit(res31greyscale_train_data, validation_data = res31greyscale_validation_data, epochs = 150, callbacks = earlyStopper)
cnnf.class_image(res31greyscale_full_data, class_names, cmap = "gray")
cnnf.model_metric_graph(history, baseline31_model, res31greyscale_validation_data, class_names= class_names,name = "baseline model for 31x31 image (normally greyscaled)")

Epoch 1/150
283/283 [==============================] - 6s 17ms/step - loss: 2.6087 - accuracy: 0.1283 - val_loss: 2.7094 - val_accuracy: 0.1067
Epoch 2/150
283/283 [==============================] - 5s 18ms/step - loss: 2.5015 - accuracy: 0.1799 - val_loss: 2.6120 - val_accuracy: 0.1570
Epoch 3/150
283/283 [==============================] - 5s 17ms/step - loss: 2.4232 - accuracy: 0.2180 - val_loss: 2.4869 - val_accuracy: 0.1710
Epoch 4/150
283/283 [==============================] - 5s 17ms/step - loss: 2.3518 - accuracy: 0.2312 - val_loss: 2.4270 - val_accuracy: 0.1720
Epoch 5/150
283/283 [==============================] - 5s 17ms/step - loss: 2.2896 - accuracy: 0.2518 - val_loss: 2.3637 - val_accuracy: 0.1890
Epoch 6/150
283/283 [==============================] - 5s 16ms/step - loss: 2.2534 - accuracy: 0.2575 - val_loss: 2.3032 - val_accuracy: 0.2070
Epoch 7/150
283/283 [==============================] - 5s 16ms/step - loss: 2.2163 - accuracy: 0.2841 - val_loss: 2.3038 - val_accuracy: 0.2103
Epoch 8/150
283/283 [==============================] - 5s 17ms/step - loss: 2.1858 - accuracy: 0.3001 - val_loss: 2.2461 - val_accuracy: 0.2330
Epoch 9/150
283/283 [==============================] - 5s 17ms/step - loss: 2.1552 - accuracy: 0.3130 - val_loss: 2.2042 - val_accuracy: 0.2780
Epoch 10/150
283/283 [==============================] - 5s 17ms/step - loss: 2.1246 - accuracy: 0.3238 - val_loss: 2.2150 - val_accuracy: 0.2727
Epoch 11/150
283/283 [==============================] - 5s 16ms/step - loss: 2.0957 - accuracy: 0.3368 - val_loss: 2.1553 - val_accuracy: 0.2747
Epoch 12/150
283/283 [==============================] - 5s 17ms/step - loss: 2.0554 - accuracy: 0.3487 - val_loss: 2.1234 - val_accuracy: 0.3200
Epoch 13/150
283/283 [==============================] - 5s 17ms/step - loss: 2.0180 - accuracy: 0.3636 - val_loss: 2.0647 - val_accuracy: 0.3263
Epoch 14/150
283/283 [==============================] - 5s 16ms/step - loss: 1.9950 - accuracy: 0.3700 - val_loss: 2.0547 - val_accuracy: 0.3450
Epoch 15/150
283/283 [==============================] - 5s 16ms/step - loss: 1.9582 - accuracy: 0.3825 - val_loss: 2.0025 - val_accuracy: 0.3643
Epoch 16/150
283/283 [==============================] - 5s 16ms/step - loss: 1.9265 - accuracy: 0.3983 - val_loss: 2.0133 - val_accuracy: 0.3367
Epoch 17/150
283/283 [==============================] - 5s 16ms/step - loss: 1.8962 - accuracy: 0.4072 - val_loss: 1.9982 - val_accuracy: 0.3367
Epoch 18/150
283/283 [==============================] - 5s 16ms/step - loss: 1.8799 - accuracy: 0.4066 - val_loss: 1.9766 - val_accuracy: 0.3503
Epoch 19/150
283/283 [==============================] - 5s 16ms/step - loss: 1.8576 - accuracy: 0.4202 - val_loss: 1.9206 - val_accuracy: 0.3940
Epoch 20/150
283/283 [==============================] - 5s 16ms/step - loss: 1.8365 - accuracy: 0.4243 - val_loss: 1.9021 - val_accuracy: 0.3980
Epoch 21/150
283/283 [==============================] - 5s 16ms/step - loss: 1.8239 - accuracy: 0.4287 - val_loss: 1.8985 - val_accuracy: 0.3957
Epoch 22/150
283/283 [==============================] - 5s 15ms/step - loss: 1.8086 - accuracy: 0.4343 - val_loss: 2.0177 - val_accuracy: 0.3470
Epoch 23/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7901 - accuracy: 0.4456 - val_loss: 1.9101 - val_accuracy: 0.3717
Epoch 24/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7761 - accuracy: 0.4472 - val_loss: 1.9037 - val_accuracy: 0.3717
Epoch 25/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7670 - accuracy: 0.4543 - val_loss: 1.8465 - val_accuracy: 0.4087
Epoch 26/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7536 - accuracy: 0.4502 - val_loss: 1.8077 - val_accuracy: 0.4460
Epoch 27/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7484 - accuracy: 0.4567 - val_loss: 1.8285 - val_accuracy: 0.4203
Epoch 28/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7325 - accuracy: 0.4681 - val_loss: 1.8090 - val_accuracy: 0.4287
Epoch 29/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7319 - accuracy: 0.4600 - val_loss: 1.7781 - val_accuracy: 0.4400
Epoch 30/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7125 - accuracy: 0.4721 - val_loss: 1.7838 - val_accuracy: 0.4480
Epoch 31/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7066 - accuracy: 0.4752 - val_loss: 1.8057 - val_accuracy: 0.4363
Epoch 32/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6919 - accuracy: 0.4788 - val_loss: 1.7779 - val_accuracy: 0.4337
Epoch 33/150
283/283 [==============================] - 5s 15ms/step - loss: 1.6917 - accuracy: 0.4753 - val_loss: 1.7909 - val_accuracy: 0.4217
Epoch 34/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6899 - accuracy: 0.4823 - val_loss: 1.7527 - val_accuracy: 0.4563
Epoch 35/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6691 - accuracy: 0.4883 - val_loss: 1.8104 - val_accuracy: 0.4203
Epoch 36/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6608 - accuracy: 0.4864 - val_loss: 1.7318 - val_accuracy: 0.4477
Epoch 37/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6540 - accuracy: 0.4896 - val_loss: 1.7157 - val_accuracy: 0.4647
Epoch 38/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6431 - accuracy: 0.4952 - val_loss: 1.7511 - val_accuracy: 0.4343
Epoch 39/150
283/283 [==============================] - 5s 15ms/step - loss: 1.6419 - accuracy: 0.4986 - val_loss: 1.7892 - val_accuracy: 0.4290
Epoch 40/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6364 - accuracy: 0.4939 - val_loss: 1.7902 - val_accuracy: 0.4263
Epoch 41/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6258 - accuracy: 0.4992 - val_loss: 1.6750 - val_accuracy: 0.4800
Epoch 42/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6255 - accuracy: 0.5002 - val_loss: 1.6711 - val_accuracy: 0.4957
Epoch 43/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6124 - accuracy: 0.5058 - val_loss: 1.6840 - val_accuracy: 0.4713
Epoch 44/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6002 - accuracy: 0.5078 - val_loss: 1.6816 - val_accuracy: 0.4607
Epoch 45/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6137 - accuracy: 0.5030 - val_loss: 1.6778 - val_accuracy: 0.4753
Epoch 46/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5950 - accuracy: 0.5090 - val_loss: 1.6333 - val_accuracy: 0.4980
Epoch 47/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5888 - accuracy: 0.5128 - val_loss: 1.6336 - val_accuracy: 0.4937
Epoch 48/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5840 - accuracy: 0.5195 - val_loss: 1.6674 - val_accuracy: 0.4807
Epoch 49/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5760 - accuracy: 0.5157 - val_loss: 1.6587 - val_accuracy: 0.4793
Epoch 50/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5711 - accuracy: 0.5164 - val_loss: 1.6311 - val_accuracy: 0.4953
Epoch 51/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5539 - accuracy: 0.5265 - val_loss: 1.6442 - val_accuracy: 0.4897
Epoch 52/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5526 - accuracy: 0.5216 - val_loss: 1.6261 - val_accuracy: 0.5063
Epoch 53/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5465 - accuracy: 0.5261 - val_loss: 1.6170 - val_accuracy: 0.4937
Epoch 54/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5444 - accuracy: 0.5234 - val_loss: 1.5920 - val_accuracy: 0.5203
Epoch 55/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5425 - accuracy: 0.5256 - val_loss: 1.5835 - val_accuracy: 0.5250
Epoch 56/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5312 - accuracy: 0.5305 - val_loss: 1.5985 - val_accuracy: 0.5043
Epoch 57/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5251 - accuracy: 0.5296 - val_loss: 1.5820 - val_accuracy: 0.5137
Epoch 58/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5224 - accuracy: 0.5329 - val_loss: 1.5914 - val_accuracy: 0.5053
Epoch 59/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5122 - accuracy: 0.5388 - val_loss: 1.5863 - val_accuracy: 0.5047
Epoch 60/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5098 - accuracy: 0.5356 - val_loss: 1.5730 - val_accuracy: 0.5207
Epoch 61/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5034 - accuracy: 0.5364 - val_loss: 1.5517 - val_accuracy: 0.5230
Epoch 62/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5033 - accuracy: 0.5407 - val_loss: 1.5497 - val_accuracy: 0.5233
Epoch 63/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4957 - accuracy: 0.5398 - val_loss: 1.5672 - val_accuracy: 0.5257
Epoch 64/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4886 - accuracy: 0.5400 - val_loss: 1.6713 - val_accuracy: 0.4773
Epoch 65/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4848 - accuracy: 0.5414 - val_loss: 1.5683 - val_accuracy: 0.5123
Epoch 66/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4815 - accuracy: 0.5443 - val_loss: 1.5558 - val_accuracy: 0.5240
Epoch 67/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4765 - accuracy: 0.5488 - val_loss: 1.5360 - val_accuracy: 0.5307
Epoch 68/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4726 - accuracy: 0.5463 - val_loss: 1.6549 - val_accuracy: 0.4770
Epoch 69/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4654 - accuracy: 0.5459 - val_loss: 1.5443 - val_accuracy: 0.5197
Epoch 70/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4566 - accuracy: 0.5577 - val_loss: 1.5394 - val_accuracy: 0.5253
Epoch 71/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4609 - accuracy: 0.5498 - val_loss: 1.5157 - val_accuracy: 0.5167
Epoch 72/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4585 - accuracy: 0.5498 - val_loss: 1.5891 - val_accuracy: 0.5100
Epoch 73/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4491 - accuracy: 0.5525 - val_loss: 1.5177 - val_accuracy: 0.5203
Epoch 74/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4459 - accuracy: 0.5573 - val_loss: 1.5231 - val_accuracy: 0.5300
Epoch 75/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4443 - accuracy: 0.5564 - val_loss: 1.5276 - val_accuracy: 0.5267
Epoch 76/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4400 - accuracy: 0.5574 - val_loss: 1.5083 - val_accuracy: 0.5317
Epoch 77/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4338 - accuracy: 0.5603 - val_loss: 1.5699 - val_accuracy: 0.5077
Epoch 78/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4376 - accuracy: 0.5583 - val_loss: 1.5679 - val_accuracy: 0.5077
Epoch 79/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4284 - accuracy: 0.5565 - val_loss: 1.4781 - val_accuracy: 0.5487
Epoch 80/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4133 - accuracy: 0.5681 - val_loss: 1.5038 - val_accuracy: 0.5290
Epoch 81/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4163 - accuracy: 0.5647 - val_loss: 1.5086 - val_accuracy: 0.5347
Epoch 82/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4130 - accuracy: 0.5648 - val_loss: 1.5596 - val_accuracy: 0.5000
Epoch 83/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4139 - accuracy: 0.5655 - val_loss: 1.5484 - val_accuracy: 0.5113
Epoch 84/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4041 - accuracy: 0.5712 - val_loss: 1.4790 - val_accuracy: 0.5407
Epoch 85/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4074 - accuracy: 0.5646 - val_loss: 1.4755 - val_accuracy: 0.5367
Epoch 86/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3997 - accuracy: 0.5647 - val_loss: 1.4636 - val_accuracy: 0.5430
Epoch 87/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3894 - accuracy: 0.5757 - val_loss: 1.5272 - val_accuracy: 0.5220
Epoch 88/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3929 - accuracy: 0.5710 - val_loss: 1.4348 - val_accuracy: 0.5507
Epoch 89/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3867 - accuracy: 0.5692 - val_loss: 1.4321 - val_accuracy: 0.5543
Epoch 90/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3881 - accuracy: 0.5753 - val_loss: 1.4614 - val_accuracy: 0.5453
Epoch 91/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3795 - accuracy: 0.5712 - val_loss: 1.4380 - val_accuracy: 0.5513
Epoch 92/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3759 - accuracy: 0.5780 - val_loss: 1.4079 - val_accuracy: 0.5580
Epoch 93/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3724 - accuracy: 0.5795 - val_loss: 1.5134 - val_accuracy: 0.5207
Epoch 94/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3683 - accuracy: 0.5771 - val_loss: 1.4150 - val_accuracy: 0.5633
Epoch 95/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3632 - accuracy: 0.5784 - val_loss: 1.4174 - val_accuracy: 0.5607
Epoch 96/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3575 - accuracy: 0.5847 - val_loss: 1.4380 - val_accuracy: 0.5477
Epoch 97/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3670 - accuracy: 0.5759 - val_loss: 1.4202 - val_accuracy: 0.5563
Epoch 98/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3530 - accuracy: 0.5885 - val_loss: 1.4821 - val_accuracy: 0.5480
Epoch 99/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3556 - accuracy: 0.5811 - val_loss: 1.3923 - val_accuracy: 0.5717
Epoch 100/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3506 - accuracy: 0.5821 - val_loss: 1.4417 - val_accuracy: 0.5447
Epoch 101/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3571 - accuracy: 0.5791 - val_loss: 1.4013 - val_accuracy: 0.5630
Epoch 102/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3499 - accuracy: 0.5847 - val_loss: 1.4781 - val_accuracy: 0.5390
Epoch 103/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3385 - accuracy: 0.5899 - val_loss: 1.4259 - val_accuracy: 0.5503
Epoch 104/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3356 - accuracy: 0.5887 - val_loss: 1.4038 - val_accuracy: 0.5563
Epoch 105/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3293 - accuracy: 0.5935 - val_loss: 1.3685 - val_accuracy: 0.5837
Epoch 106/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3261 - accuracy: 0.5916 - val_loss: 1.4137 - val_accuracy: 0.5567
Epoch 107/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3272 - accuracy: 0.5930 - val_loss: 1.3865 - val_accuracy: 0.5740
Epoch 108/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3176 - accuracy: 0.5935 - val_loss: 1.5122 - val_accuracy: 0.5290
Epoch 109/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3202 - accuracy: 0.5969 - val_loss: 1.4436 - val_accuracy: 0.5400
Epoch 110/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3277 - accuracy: 0.5888 - val_loss: 1.4327 - val_accuracy: 0.5460
Epoch 111/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3160 - accuracy: 0.5912 - val_loss: 1.3982 - val_accuracy: 0.5610
Epoch 112/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3211 - accuracy: 0.5955 - val_loss: 1.4000 - val_accuracy: 0.5500
Epoch 113/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3125 - accuracy: 0.5955 - val_loss: 1.5424 - val_accuracy: 0.5090
Epoch 114/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3120 - accuracy: 0.5965 - val_loss: 1.3906 - val_accuracy: 0.5600
Epoch 115/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3001 - accuracy: 0.5996 - val_loss: 1.4020 - val_accuracy: 0.5623
Epoch 116/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2993 - accuracy: 0.5981 - val_loss: 1.4299 - val_accuracy: 0.5460
Epoch 117/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3001 - accuracy: 0.5998 - val_loss: 1.4500 - val_accuracy: 0.5437
Epoch 118/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2970 - accuracy: 0.5995 - val_loss: 1.3762 - val_accuracy: 0.5657
Epoch 119/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3047 - accuracy: 0.5943 - val_loss: 1.4198 - val_accuracy: 0.5477
Epoch 120/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2911 - accuracy: 0.6006 - val_loss: 1.3959 - val_accuracy: 0.5517
Epoch 121/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2860 - accuracy: 0.6057 - val_loss: 1.3949 - val_accuracy: 0.5623
Epoch 122/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2855 - accuracy: 0.6008 - val_loss: 1.4563 - val_accuracy: 0.5350
Epoch 123/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2889 - accuracy: 0.6005 - val_loss: 1.3314 - val_accuracy: 0.5890
Epoch 124/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2798 - accuracy: 0.6047 - val_loss: 1.4215 - val_accuracy: 0.5487
Epoch 125/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2863 - accuracy: 0.6028 - val_loss: 1.3136 - val_accuracy: 0.5873
Epoch 126/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2789 - accuracy: 0.6062 - val_loss: 1.4181 - val_accuracy: 0.5587
Epoch 127/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2767 - accuracy: 0.6091 - val_loss: 1.3925 - val_accuracy: 0.5563
Epoch 128/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2722 - accuracy: 0.6086 - val_loss: 1.3293 - val_accuracy: 0.5833
Epoch 129/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2790 - accuracy: 0.6005 - val_loss: 1.3498 - val_accuracy: 0.5820
Epoch 130/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2665 - accuracy: 0.6082 - val_loss: 1.3468 - val_accuracy: 0.5703
Epoch 131/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2647 - accuracy: 0.6073 - val_loss: 1.3768 - val_accuracy: 0.5607
Epoch 132/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2608 - accuracy: 0.6100 - val_loss: 1.3159 - val_accuracy: 0.5957
Epoch 133/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2586 - accuracy: 0.6115 - val_loss: 1.3180 - val_accuracy: 0.5933
Epoch 134/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2551 - accuracy: 0.6108 - val_loss: 1.3253 - val_accuracy: 0.5780
Epoch 135/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2529 - accuracy: 0.6128 - val_loss: 1.2984 - val_accuracy: 0.5963
Epoch 136/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2546 - accuracy: 0.6131 - val_loss: 1.3657 - val_accuracy: 0.5690
Epoch 137/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2525 - accuracy: 0.6150 - val_loss: 1.3358 - val_accuracy: 0.5740
Epoch 138/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2478 - accuracy: 0.6175 - val_loss: 1.2823 - val_accuracy: 0.6043
Epoch 139/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2464 - accuracy: 0.6135 - val_loss: 1.3091 - val_accuracy: 0.5920
Epoch 140/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2476 - accuracy: 0.6153 - val_loss: 1.3220 - val_accuracy: 0.5820
Epoch 141/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2366 - accuracy: 0.6160 - val_loss: 1.3006 - val_accuracy: 0.5887
Epoch 142/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2420 - accuracy: 0.6154 - val_loss: 1.2842 - val_accuracy: 0.6020
Epoch 143/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2376 - accuracy: 0.6184 - val_loss: 1.2658 - val_accuracy: 0.6073
Epoch 144/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2449 - accuracy: 0.6152 - val_loss: 1.3947 - val_accuracy: 0.5573
Epoch 145/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2322 - accuracy: 0.6155 - val_loss: 1.3416 - val_accuracy: 0.5817
Epoch 146/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2248 - accuracy: 0.6165 - val_loss: 1.3243 - val_accuracy: 0.5817
Epoch 147/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2296 - accuracy: 0.6172 - val_loss: 1.2479 - val_accuracy: 0.6117
Epoch 148/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2267 - accuracy: 0.6203 - val_loss: 1.2624 - val_accuracy: 0.6163
Epoch 149/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2231 - accuracy: 0.6196 - val_loss: 1.3174 - val_accuracy: 0.5803
Epoch 150/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2162 - accuracy: 0.6185 - val_loss: 1.3386 - val_accuracy: 0.5730

history = baseline31luminosity_model.fit(res31luminosity_train_data, validation_data = res31luminosity_validation_data, epochs = 150, callbacks = earlyStopper)
cnnf.class_image(res31luminosity_full_data, class_names, cmap = "gray")
cnnf.model_metric_graph(history, baseline31luminosity_model, res31luminosity_validation_data, class_names= class_names,name = "baseline model for 31x31 image (greyscaled using luminosity method)")

Epoch 1/150
283/283 [==============================] - 5s 17ms/step - loss: 2.6063 - accuracy: 0.1224 - val_loss: 2.7064 - val_accuracy: 0.0887
Epoch 2/150
283/283 [==============================] - 5s 17ms/step - loss: 2.5028 - accuracy: 0.1694 - val_loss: 2.6157 - val_accuracy: 0.1403
Epoch 3/150
283/283 [==============================] - 5s 16ms/step - loss: 2.4363 - accuracy: 0.1951 - val_loss: 2.5341 - val_accuracy: 0.1583
Epoch 4/150
283/283 [==============================] - 5s 16ms/step - loss: 2.3626 - accuracy: 0.2148 - val_loss: 2.4281 - val_accuracy: 0.1800
Epoch 5/150
283/283 [==============================] - 5s 16ms/step - loss: 2.2896 - accuracy: 0.2312 - val_loss: 2.4209 - val_accuracy: 0.1653
Epoch 6/150
283/283 [==============================] - 5s 16ms/step - loss: 2.2509 - accuracy: 0.2455 - val_loss: 2.2842 - val_accuracy: 0.1897
Epoch 7/150
283/283 [==============================] - 5s 17ms/step - loss: 2.2125 - accuracy: 0.2550 - val_loss: 2.2573 - val_accuracy: 0.2180
Epoch 8/150
283/283 [==============================] - 5s 17ms/step - loss: 2.1848 - accuracy: 0.2724 - val_loss: 2.2667 - val_accuracy: 0.2220
Epoch 9/150
283/283 [==============================] - 5s 17ms/step - loss: 2.1559 - accuracy: 0.2868 - val_loss: 2.1842 - val_accuracy: 0.2447
Epoch 10/150
283/283 [==============================] - 5s 17ms/step - loss: 2.1314 - accuracy: 0.2957 - val_loss: 2.1990 - val_accuracy: 0.2530
Epoch 11/150
283/283 [==============================] - 5s 17ms/step - loss: 2.1034 - accuracy: 0.3127 - val_loss: 2.1321 - val_accuracy: 0.2900
Epoch 12/150
283/283 [==============================] - 5s 17ms/step - loss: 2.0757 - accuracy: 0.3291 - val_loss: 2.1684 - val_accuracy: 0.2723
Epoch 13/150
283/283 [==============================] - 5s 17ms/step - loss: 2.0557 - accuracy: 0.3382 - val_loss: 2.2160 - val_accuracy: 0.2803
Epoch 14/150
283/283 [==============================] - 5s 17ms/step - loss: 2.0358 - accuracy: 0.3487 - val_loss: 2.1231 - val_accuracy: 0.2893
Epoch 15/150
283/283 [==============================] - 5s 16ms/step - loss: 2.0184 - accuracy: 0.3611 - val_loss: 2.0848 - val_accuracy: 0.3327
Epoch 16/150
283/283 [==============================] - 5s 17ms/step - loss: 1.9924 - accuracy: 0.3736 - val_loss: 2.0386 - val_accuracy: 0.3367
Epoch 17/150
283/283 [==============================] - 5s 16ms/step - loss: 1.9723 - accuracy: 0.3805 - val_loss: 2.0254 - val_accuracy: 0.3260
Epoch 18/150
283/283 [==============================] - 5s 17ms/step - loss: 1.9552 - accuracy: 0.3826 - val_loss: 2.0397 - val_accuracy: 0.3317
Epoch 19/150
283/283 [==============================] - 5s 16ms/step - loss: 1.9360 - accuracy: 0.3893 - val_loss: 2.0688 - val_accuracy: 0.3373
Epoch 20/150
283/283 [==============================] - 5s 16ms/step - loss: 1.9181 - accuracy: 0.3985 - val_loss: 1.9631 - val_accuracy: 0.3843
Epoch 21/150
283/283 [==============================] - 5s 17ms/step - loss: 1.9002 - accuracy: 0.4080 - val_loss: 2.0367 - val_accuracy: 0.3423
Epoch 22/150
283/283 [==============================] - 5s 17ms/step - loss: 1.8802 - accuracy: 0.4146 - val_loss: 1.9451 - val_accuracy: 0.3590
Epoch 23/150
283/283 [==============================] - 5s 16ms/step - loss: 1.8603 - accuracy: 0.4251 - val_loss: 1.8835 - val_accuracy: 0.3950
Epoch 24/150
283/283 [==============================] - 5s 17ms/step - loss: 1.8391 - accuracy: 0.4331 - val_loss: 2.0038 - val_accuracy: 0.3477
Epoch 25/150
283/283 [==============================] - 5s 16ms/step - loss: 1.8276 - accuracy: 0.4341 - val_loss: 1.9153 - val_accuracy: 0.3927
Epoch 26/150
283/283 [==============================] - 5s 17ms/step - loss: 1.8031 - accuracy: 0.4421 - val_loss: 1.8470 - val_accuracy: 0.4197
Epoch 27/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7879 - accuracy: 0.4466 - val_loss: 1.8781 - val_accuracy: 0.4130
Epoch 28/150
283/283 [==============================] - 5s 17ms/step - loss: 1.7720 - accuracy: 0.4541 - val_loss: 1.8665 - val_accuracy: 0.4037
Epoch 29/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7549 - accuracy: 0.4598 - val_loss: 1.8055 - val_accuracy: 0.4380
Epoch 30/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7379 - accuracy: 0.4682 - val_loss: 1.8561 - val_accuracy: 0.4157
Epoch 31/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7312 - accuracy: 0.4647 - val_loss: 1.8467 - val_accuracy: 0.4053
Epoch 32/150
283/283 [==============================] - 5s 17ms/step - loss: 1.7161 - accuracy: 0.4715 - val_loss: 1.7961 - val_accuracy: 0.4210
Epoch 33/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7030 - accuracy: 0.4754 - val_loss: 1.7741 - val_accuracy: 0.4437
Epoch 34/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6863 - accuracy: 0.4730 - val_loss: 1.7565 - val_accuracy: 0.4577
Epoch 35/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6876 - accuracy: 0.4750 - val_loss: 1.7890 - val_accuracy: 0.4283
Epoch 36/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6768 - accuracy: 0.4826 - val_loss: 1.7582 - val_accuracy: 0.4450
Epoch 37/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6627 - accuracy: 0.4867 - val_loss: 1.7538 - val_accuracy: 0.4487
Epoch 38/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6499 - accuracy: 0.4896 - val_loss: 1.7765 - val_accuracy: 0.4407
Epoch 39/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6397 - accuracy: 0.4975 - val_loss: 1.7106 - val_accuracy: 0.4563
Epoch 40/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6334 - accuracy: 0.4976 - val_loss: 1.6866 - val_accuracy: 0.4717
Epoch 41/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6239 - accuracy: 0.4980 - val_loss: 1.6846 - val_accuracy: 0.4733
Epoch 42/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6240 - accuracy: 0.4944 - val_loss: 1.6955 - val_accuracy: 0.4837
Epoch 43/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6092 - accuracy: 0.4998 - val_loss: 1.7516 - val_accuracy: 0.4473
Epoch 44/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6010 - accuracy: 0.5084 - val_loss: 1.6539 - val_accuracy: 0.4833
Epoch 45/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5951 - accuracy: 0.5078 - val_loss: 1.6856 - val_accuracy: 0.4770
Epoch 46/150
283/283 [==============================] - 5s 17ms/step - loss: 1.5920 - accuracy: 0.5119 - val_loss: 1.6379 - val_accuracy: 0.4807
Epoch 47/150
283/283 [==============================] - 5s 17ms/step - loss: 1.5848 - accuracy: 0.5090 - val_loss: 1.6709 - val_accuracy: 0.4770
Epoch 48/150
283/283 [==============================] - 5s 17ms/step - loss: 1.5782 - accuracy: 0.5078 - val_loss: 1.6400 - val_accuracy: 0.4830
Epoch 49/150
283/283 [==============================] - 5s 17ms/step - loss: 1.5727 - accuracy: 0.5101 - val_loss: 1.6595 - val_accuracy: 0.4860
Epoch 50/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5651 - accuracy: 0.5167 - val_loss: 1.6012 - val_accuracy: 0.5020
Epoch 51/150
283/283 [==============================] - 5s 17ms/step - loss: 1.5524 - accuracy: 0.5214 - val_loss: 1.6126 - val_accuracy: 0.4930
Epoch 52/150
283/283 [==============================] - 5s 17ms/step - loss: 1.5445 - accuracy: 0.5249 - val_loss: 1.6313 - val_accuracy: 0.4940
Epoch 53/150
283/283 [==============================] - 5s 17ms/step - loss: 1.5446 - accuracy: 0.5232 - val_loss: 1.5938 - val_accuracy: 0.5047
Epoch 54/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5408 - accuracy: 0.5207 - val_loss: 1.6513 - val_accuracy: 0.4727
Epoch 55/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5284 - accuracy: 0.5266 - val_loss: 1.5955 - val_accuracy: 0.5010
Epoch 56/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5226 - accuracy: 0.5265 - val_loss: 1.6530 - val_accuracy: 0.4763
Epoch 57/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5287 - accuracy: 0.5254 - val_loss: 1.5674 - val_accuracy: 0.5117
Epoch 58/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5216 - accuracy: 0.5328 - val_loss: 1.6385 - val_accuracy: 0.4830
Epoch 59/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5190 - accuracy: 0.5332 - val_loss: 1.5619 - val_accuracy: 0.5093
Epoch 60/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5066 - accuracy: 0.5312 - val_loss: 1.5612 - val_accuracy: 0.5170
Epoch 61/150
283/283 [==============================] - 5s 17ms/step - loss: 1.5029 - accuracy: 0.5327 - val_loss: 1.5763 - val_accuracy: 0.5107
Epoch 62/150
283/283 [==============================] - 5s 17ms/step - loss: 1.5049 - accuracy: 0.5290 - val_loss: 1.5507 - val_accuracy: 0.5157
Epoch 63/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4920 - accuracy: 0.5379 - val_loss: 1.5690 - val_accuracy: 0.5170
Epoch 64/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4970 - accuracy: 0.5353 - val_loss: 1.5696 - val_accuracy: 0.5057
Epoch 65/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4844 - accuracy: 0.5433 - val_loss: 1.6313 - val_accuracy: 0.4803
Epoch 66/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4754 - accuracy: 0.5401 - val_loss: 1.5134 - val_accuracy: 0.5340
Epoch 67/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4852 - accuracy: 0.5407 - val_loss: 1.5626 - val_accuracy: 0.5203
Epoch 68/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4709 - accuracy: 0.5449 - val_loss: 1.4907 - val_accuracy: 0.5453
Epoch 69/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4602 - accuracy: 0.5525 - val_loss: 1.6792 - val_accuracy: 0.4683
Epoch 70/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4569 - accuracy: 0.5484 - val_loss: 1.6204 - val_accuracy: 0.4880
Epoch 71/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4628 - accuracy: 0.5436 - val_loss: 1.7631 - val_accuracy: 0.4347
Epoch 72/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4589 - accuracy: 0.5482 - val_loss: 1.5313 - val_accuracy: 0.5217
Epoch 73/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4501 - accuracy: 0.5491 - val_loss: 1.5195 - val_accuracy: 0.5290
Epoch 74/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4433 - accuracy: 0.5572 - val_loss: 1.4911 - val_accuracy: 0.5353
Epoch 75/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4414 - accuracy: 0.5521 - val_loss: 1.4999 - val_accuracy: 0.5317
Epoch 76/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4333 - accuracy: 0.5548 - val_loss: 1.4409 - val_accuracy: 0.5663
Epoch 77/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4336 - accuracy: 0.5544 - val_loss: 1.6517 - val_accuracy: 0.4847
Epoch 78/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4335 - accuracy: 0.5609 - val_loss: 1.4710 - val_accuracy: 0.5537
Epoch 79/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4222 - accuracy: 0.5595 - val_loss: 1.5298 - val_accuracy: 0.5260
Epoch 80/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4119 - accuracy: 0.5647 - val_loss: 1.4650 - val_accuracy: 0.5450
Epoch 81/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4157 - accuracy: 0.5562 - val_loss: 1.4640 - val_accuracy: 0.5470
Epoch 82/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4094 - accuracy: 0.5605 - val_loss: 1.4421 - val_accuracy: 0.5540
Epoch 83/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4065 - accuracy: 0.5649 - val_loss: 1.4964 - val_accuracy: 0.5330
Epoch 84/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4005 - accuracy: 0.5656 - val_loss: 1.4465 - val_accuracy: 0.5380
Epoch 85/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3962 - accuracy: 0.5671 - val_loss: 1.4374 - val_accuracy: 0.5463
Epoch 86/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3880 - accuracy: 0.5693 - val_loss: 1.5104 - val_accuracy: 0.5217
Epoch 87/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3904 - accuracy: 0.5703 - val_loss: 1.5098 - val_accuracy: 0.5243
Epoch 88/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3819 - accuracy: 0.5696 - val_loss: 1.4578 - val_accuracy: 0.5443
Epoch 89/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3796 - accuracy: 0.5735 - val_loss: 1.5387 - val_accuracy: 0.5147
Epoch 90/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3793 - accuracy: 0.5719 - val_loss: 1.5192 - val_accuracy: 0.4980
Epoch 91/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3730 - accuracy: 0.5706 - val_loss: 1.4750 - val_accuracy: 0.5350
Epoch 92/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3670 - accuracy: 0.5776 - val_loss: 1.4148 - val_accuracy: 0.5577
Epoch 93/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3601 - accuracy: 0.5772 - val_loss: 1.4200 - val_accuracy: 0.5517
Epoch 94/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3594 - accuracy: 0.5770 - val_loss: 1.4051 - val_accuracy: 0.5637
Epoch 95/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3575 - accuracy: 0.5810 - val_loss: 1.4740 - val_accuracy: 0.5370
Epoch 96/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3640 - accuracy: 0.5769 - val_loss: 1.3730 - val_accuracy: 0.5810
Epoch 97/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3452 - accuracy: 0.5820 - val_loss: 1.4559 - val_accuracy: 0.5380
Epoch 98/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3538 - accuracy: 0.5801 - val_loss: 1.4350 - val_accuracy: 0.5480
Epoch 99/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3437 - accuracy: 0.5850 - val_loss: 1.3673 - val_accuracy: 0.5780
Epoch 100/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3308 - accuracy: 0.5844 - val_loss: 1.4135 - val_accuracy: 0.5523
Epoch 101/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3352 - accuracy: 0.5843 - val_loss: 1.3741 - val_accuracy: 0.5730
Epoch 102/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3293 - accuracy: 0.5875 - val_loss: 1.3620 - val_accuracy: 0.5767
Epoch 103/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3332 - accuracy: 0.5853 - val_loss: 1.3590 - val_accuracy: 0.5877
Epoch 104/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3268 - accuracy: 0.5856 - val_loss: 1.3910 - val_accuracy: 0.5603
Epoch 105/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3198 - accuracy: 0.5913 - val_loss: 1.3704 - val_accuracy: 0.5770
Epoch 106/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3166 - accuracy: 0.5908 - val_loss: 1.3862 - val_accuracy: 0.5690
Epoch 107/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3147 - accuracy: 0.5909 - val_loss: 1.3583 - val_accuracy: 0.5770
Epoch 108/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3164 - accuracy: 0.5906 - val_loss: 1.3342 - val_accuracy: 0.5917
Epoch 109/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3112 - accuracy: 0.5960 - val_loss: 1.3450 - val_accuracy: 0.5843
Epoch 110/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2945 - accuracy: 0.5990 - val_loss: 1.3374 - val_accuracy: 0.5800
Epoch 111/150
283/283 [==============================] - 6s 18ms/step - loss: 1.3016 - accuracy: 0.5978 - val_loss: 1.3935 - val_accuracy: 0.5557
Epoch 112/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2910 - accuracy: 0.5971 - val_loss: 1.3175 - val_accuracy: 0.5907
Epoch 113/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2876 - accuracy: 0.6031 - val_loss: 1.3329 - val_accuracy: 0.5883
Epoch 114/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2867 - accuracy: 0.5994 - val_loss: 1.5039 - val_accuracy: 0.5217
Epoch 115/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2820 - accuracy: 0.6079 - val_loss: 1.3679 - val_accuracy: 0.5710
Epoch 116/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2780 - accuracy: 0.6019 - val_loss: 1.3559 - val_accuracy: 0.5800
Epoch 117/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2770 - accuracy: 0.6076 - val_loss: 1.3213 - val_accuracy: 0.5957
Epoch 118/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2730 - accuracy: 0.6080 - val_loss: 1.3322 - val_accuracy: 0.5767
Epoch 119/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2645 - accuracy: 0.6067 - val_loss: 1.3369 - val_accuracy: 0.5750
Epoch 120/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2706 - accuracy: 0.6069 - val_loss: 1.4067 - val_accuracy: 0.5510
Epoch 121/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2616 - accuracy: 0.6071 - val_loss: 1.3108 - val_accuracy: 0.6003
Epoch 122/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2610 - accuracy: 0.6068 - val_loss: 1.2948 - val_accuracy: 0.5987
Epoch 123/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2565 - accuracy: 0.6092 - val_loss: 1.3475 - val_accuracy: 0.5733
Epoch 124/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2490 - accuracy: 0.6122 - val_loss: 1.3091 - val_accuracy: 0.6000
Epoch 125/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2465 - accuracy: 0.6141 - val_loss: 1.3190 - val_accuracy: 0.5827
Epoch 126/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2438 - accuracy: 0.6131 - val_loss: 1.3704 - val_accuracy: 0.5707
Epoch 127/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2467 - accuracy: 0.6139 - val_loss: 1.3489 - val_accuracy: 0.5627
Epoch 128/150
283/283 [==============================] - 6s 18ms/step - loss: 1.2378 - accuracy: 0.6180 - val_loss: 1.3526 - val_accuracy: 0.5790
Epoch 129/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2410 - accuracy: 0.6139 - val_loss: 1.2705 - val_accuracy: 0.6083
Epoch 130/150
283/283 [==============================] - 6s 18ms/step - loss: 1.2297 - accuracy: 0.6171 - val_loss: 1.2404 - val_accuracy: 0.6240
Epoch 131/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2261 - accuracy: 0.6225 - val_loss: 1.3065 - val_accuracy: 0.5933
Epoch 132/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2286 - accuracy: 0.6182 - val_loss: 1.3260 - val_accuracy: 0.5840
Epoch 133/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2286 - accuracy: 0.6218 - val_loss: 1.2427 - val_accuracy: 0.6243
Epoch 134/150
283/283 [==============================] - 6s 18ms/step - loss: 1.2188 - accuracy: 0.6196 - val_loss: 1.2618 - val_accuracy: 0.6163
Epoch 135/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2173 - accuracy: 0.6222 - val_loss: 1.2499 - val_accuracy: 0.6213
Epoch 136/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2170 - accuracy: 0.6225 - val_loss: 1.2753 - val_accuracy: 0.6100
Epoch 137/150
283/283 [==============================] - 6s 18ms/step - loss: 1.2134 - accuracy: 0.6272 - val_loss: 1.2345 - val_accuracy: 0.6247
Epoch 138/150
283/283 [==============================] - 6s 19ms/step - loss: 1.2028 - accuracy: 0.6298 - val_loss: 1.3371 - val_accuracy: 0.5710
Epoch 139/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2129 - accuracy: 0.6266 - val_loss: 1.2382 - val_accuracy: 0.6247
Epoch 140/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2101 - accuracy: 0.6297 - val_loss: 1.2550 - val_accuracy: 0.6210
Epoch 141/150
283/283 [==============================] - 5s 17ms/step - loss: 1.1998 - accuracy: 0.6305 - val_loss: 1.2408 - val_accuracy: 0.6243
Epoch 142/150
283/283 [==============================] - 5s 17ms/step - loss: 1.1921 - accuracy: 0.6331 - val_loss: 1.2618 - val_accuracy: 0.6120
Epoch 143/150
283/283 [==============================] - 5s 17ms/step - loss: 1.1885 - accuracy: 0.6321 - val_loss: 1.2414 - val_accuracy: 0.6213
Epoch 144/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2065 - accuracy: 0.6313 - val_loss: 1.2485 - val_accuracy: 0.6143
Epoch 145/150
283/283 [==============================] - 5s 17ms/step - loss: 1.1859 - accuracy: 0.6326 - val_loss: 1.2172 - val_accuracy: 0.6343
Epoch 146/150
283/283 [==============================] - 5s 17ms/step - loss: 1.1878 - accuracy: 0.6313 - val_loss: 1.2465 - val_accuracy: 0.6173
Epoch 147/150
283/283 [==============================] - 5s 17ms/step - loss: 1.1842 - accuracy: 0.6368 - val_loss: 1.2771 - val_accuracy: 0.5970
Epoch 148/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1839 - accuracy: 0.6355 - val_loss: 1.2203 - val_accuracy: 0.6243
Epoch 149/150
283/283 [==============================] - 5s 17ms/step - loss: 1.1832 - accuracy: 0.6368 - val_loss: 1.2591 - val_accuracy: 0.6160
Epoch 150/150
283/283 [==============================] - 5s 17ms/step - loss: 1.1706 - accuracy: 0.6450 - val_loss: 1.2177 - val_accuracy: 0.6267

history = baseline31desaturation_model.fit(res31desaturation_train_data, validation_data = res31desaturation_validation_data, epochs = 150, callbacks = earlyStopper)
cnnf.class_image(res31desaturation_full_data, class_names, cmap = "gray")
cnnf.model_metric_graph(history, baseline31desaturation_model, res31desaturation_validation_data, class_names= class_names,name = "baseline model for 31x31 image (greyscaled using desaturation method)")

Epoch 1/150
283/283 [==============================] - 5s 17ms/step - loss: 2.6026 - accuracy: 0.1380 - val_loss: 2.7011 - val_accuracy: 0.1083
Epoch 2/150
283/283 [==============================] - 5s 16ms/step - loss: 2.4533 - accuracy: 0.2009 - val_loss: 2.5655 - val_accuracy: 0.1530
Epoch 3/150
283/283 [==============================] - 5s 16ms/step - loss: 2.3437 - accuracy: 0.2509 - val_loss: 2.4339 - val_accuracy: 0.1813
Epoch 4/150
283/283 [==============================] - 5s 17ms/step - loss: 2.2562 - accuracy: 0.2829 - val_loss: 2.3488 - val_accuracy: 0.2227
Epoch 5/150
283/283 [==============================] - 5s 16ms/step - loss: 2.1928 - accuracy: 0.3139 - val_loss: 2.3247 - val_accuracy: 0.2287
Epoch 6/150
283/283 [==============================] - 5s 16ms/step - loss: 2.1545 - accuracy: 0.3293 - val_loss: 2.2466 - val_accuracy: 0.2720
Epoch 7/150
283/283 [==============================] - 5s 16ms/step - loss: 2.1193 - accuracy: 0.3404 - val_loss: 2.2192 - val_accuracy: 0.2880
Epoch 8/150
283/283 [==============================] - 5s 16ms/step - loss: 2.0893 - accuracy: 0.3552 - val_loss: 2.1857 - val_accuracy: 0.3023
Epoch 9/150
283/283 [==============================] - 5s 16ms/step - loss: 2.0647 - accuracy: 0.3656 - val_loss: 2.1673 - val_accuracy: 0.3177
Epoch 10/150
283/283 [==============================] - 5s 16ms/step - loss: 2.0353 - accuracy: 0.3756 - val_loss: 2.1951 - val_accuracy: 0.2970
Epoch 11/150
283/283 [==============================] - 5s 16ms/step - loss: 2.0041 - accuracy: 0.3901 - val_loss: 2.1045 - val_accuracy: 0.3343
Epoch 12/150
283/283 [==============================] - 5s 16ms/step - loss: 1.9805 - accuracy: 0.3964 - val_loss: 2.0605 - val_accuracy: 0.3390
Epoch 13/150
283/283 [==============================] - 5s 16ms/step - loss: 1.9566 - accuracy: 0.4026 - val_loss: 2.0713 - val_accuracy: 0.3363
Epoch 14/150
283/283 [==============================] - 5s 16ms/step - loss: 1.9341 - accuracy: 0.4087 - val_loss: 2.0426 - val_accuracy: 0.3437
Epoch 15/150
283/283 [==============================] - 5s 16ms/step - loss: 1.9124 - accuracy: 0.4174 - val_loss: 2.0489 - val_accuracy: 0.3523
Epoch 16/150
283/283 [==============================] - 5s 16ms/step - loss: 1.8878 - accuracy: 0.4219 - val_loss: 2.0230 - val_accuracy: 0.3553
Epoch 17/150
283/283 [==============================] - 5s 16ms/step - loss: 1.8698 - accuracy: 0.4322 - val_loss: 1.9688 - val_accuracy: 0.3703
Epoch 18/150
283/283 [==============================] - 5s 16ms/step - loss: 1.8569 - accuracy: 0.4308 - val_loss: 2.0094 - val_accuracy: 0.3647
Epoch 19/150
283/283 [==============================] - 5s 16ms/step - loss: 1.8357 - accuracy: 0.4412 - val_loss: 1.9415 - val_accuracy: 0.3810
Epoch 20/150
283/283 [==============================] - 5s 16ms/step - loss: 1.8217 - accuracy: 0.4456 - val_loss: 1.9150 - val_accuracy: 0.3820
Epoch 21/150
283/283 [==============================] - 5s 16ms/step - loss: 1.8157 - accuracy: 0.4445 - val_loss: 1.9028 - val_accuracy: 0.3963
Epoch 22/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7920 - accuracy: 0.4527 - val_loss: 2.0439 - val_accuracy: 0.3610
Epoch 23/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7797 - accuracy: 0.4582 - val_loss: 1.8999 - val_accuracy: 0.3983
Epoch 24/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7636 - accuracy: 0.4569 - val_loss: 1.8940 - val_accuracy: 0.3930
Epoch 25/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7519 - accuracy: 0.4638 - val_loss: 1.8371 - val_accuracy: 0.4130
Epoch 26/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7378 - accuracy: 0.4700 - val_loss: 1.9151 - val_accuracy: 0.3820
Epoch 27/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7268 - accuracy: 0.4709 - val_loss: 1.8069 - val_accuracy: 0.4230
Epoch 28/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7170 - accuracy: 0.4753 - val_loss: 1.8362 - val_accuracy: 0.4173
Epoch 29/150
283/283 [==============================] - 5s 16ms/step - loss: 1.7030 - accuracy: 0.4787 - val_loss: 1.9090 - val_accuracy: 0.3910
Epoch 30/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6920 - accuracy: 0.4783 - val_loss: 1.8621 - val_accuracy: 0.3943
Epoch 31/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6859 - accuracy: 0.4837 - val_loss: 1.8061 - val_accuracy: 0.4153
Epoch 32/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6731 - accuracy: 0.4854 - val_loss: 1.8779 - val_accuracy: 0.4097
Epoch 33/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6557 - accuracy: 0.4900 - val_loss: 1.7849 - val_accuracy: 0.4140
Epoch 34/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6480 - accuracy: 0.4960 - val_loss: 1.7890 - val_accuracy: 0.4263
Epoch 35/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6363 - accuracy: 0.4994 - val_loss: 1.7354 - val_accuracy: 0.4417
Epoch 36/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6270 - accuracy: 0.5029 - val_loss: 1.7762 - val_accuracy: 0.4267
Epoch 37/150
283/283 [==============================] - 5s 16ms/step - loss: 1.6201 - accuracy: 0.5027 - val_loss: 1.7330 - val_accuracy: 0.4363
Epoch 38/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6088 - accuracy: 0.5062 - val_loss: 1.6878 - val_accuracy: 0.4510
Epoch 39/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5995 - accuracy: 0.5063 - val_loss: 1.7255 - val_accuracy: 0.4440
Epoch 40/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5904 - accuracy: 0.5085 - val_loss: 1.7136 - val_accuracy: 0.4417
Epoch 41/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5763 - accuracy: 0.5097 - val_loss: 1.6858 - val_accuracy: 0.4607
Epoch 42/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5713 - accuracy: 0.5103 - val_loss: 1.7008 - val_accuracy: 0.4567
Epoch 43/150
283/283 [==============================] - 5s 17ms/step - loss: 1.5692 - accuracy: 0.5184 - val_loss: 1.6497 - val_accuracy: 0.4677
Epoch 44/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5517 - accuracy: 0.5219 - val_loss: 1.6311 - val_accuracy: 0.4720
Epoch 45/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5527 - accuracy: 0.5246 - val_loss: 1.6732 - val_accuracy: 0.4510
Epoch 46/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5391 - accuracy: 0.5250 - val_loss: 1.6500 - val_accuracy: 0.4683
Epoch 47/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5321 - accuracy: 0.5233 - val_loss: 1.6368 - val_accuracy: 0.4667
Epoch 48/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5248 - accuracy: 0.5223 - val_loss: 1.6478 - val_accuracy: 0.4557
Epoch 49/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5117 - accuracy: 0.5302 - val_loss: 1.6446 - val_accuracy: 0.4613
Epoch 50/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5088 - accuracy: 0.5301 - val_loss: 1.6117 - val_accuracy: 0.4750
Epoch 51/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5091 - accuracy: 0.5308 - val_loss: 1.6862 - val_accuracy: 0.4490
Epoch 52/150
283/283 [==============================] - 5s 17ms/step - loss: 1.5041 - accuracy: 0.5279 - val_loss: 1.5831 - val_accuracy: 0.4850
Epoch 53/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4848 - accuracy: 0.5336 - val_loss: 1.5944 - val_accuracy: 0.4680
Epoch 54/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4867 - accuracy: 0.5360 - val_loss: 1.6392 - val_accuracy: 0.4557
Epoch 55/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4811 - accuracy: 0.5392 - val_loss: 1.5833 - val_accuracy: 0.4853
Epoch 56/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4691 - accuracy: 0.5413 - val_loss: 1.5928 - val_accuracy: 0.4830
Epoch 57/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4702 - accuracy: 0.5404 - val_loss: 1.5638 - val_accuracy: 0.4930
Epoch 58/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4524 - accuracy: 0.5445 - val_loss: 1.5344 - val_accuracy: 0.5030
Epoch 59/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4523 - accuracy: 0.5481 - val_loss: 1.5436 - val_accuracy: 0.4917
Epoch 60/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4481 - accuracy: 0.5456 - val_loss: 1.5158 - val_accuracy: 0.5060
Epoch 61/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4390 - accuracy: 0.5487 - val_loss: 1.5031 - val_accuracy: 0.5110
Epoch 62/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4427 - accuracy: 0.5477 - val_loss: 1.5658 - val_accuracy: 0.4977
Epoch 63/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4306 - accuracy: 0.5546 - val_loss: 1.5479 - val_accuracy: 0.4950
Epoch 64/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4250 - accuracy: 0.5562 - val_loss: 1.5128 - val_accuracy: 0.4997
Epoch 65/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4214 - accuracy: 0.5554 - val_loss: 1.5263 - val_accuracy: 0.4950
Epoch 66/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4192 - accuracy: 0.5551 - val_loss: 1.6112 - val_accuracy: 0.4750
Epoch 67/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4167 - accuracy: 0.5570 - val_loss: 1.5227 - val_accuracy: 0.5030
Epoch 68/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4054 - accuracy: 0.5587 - val_loss: 1.4898 - val_accuracy: 0.5097
Epoch 69/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4002 - accuracy: 0.5595 - val_loss: 1.4705 - val_accuracy: 0.5277
Epoch 70/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3988 - accuracy: 0.5598 - val_loss: 1.4556 - val_accuracy: 0.5297
Epoch 71/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3958 - accuracy: 0.5644 - val_loss: 1.4928 - val_accuracy: 0.5063
Epoch 72/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3868 - accuracy: 0.5642 - val_loss: 1.4783 - val_accuracy: 0.5167
Epoch 73/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3801 - accuracy: 0.5644 - val_loss: 1.4413 - val_accuracy: 0.5387
Epoch 74/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3788 - accuracy: 0.5690 - val_loss: 1.5009 - val_accuracy: 0.5113
Epoch 75/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3700 - accuracy: 0.5677 - val_loss: 1.4553 - val_accuracy: 0.5180
Epoch 76/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3626 - accuracy: 0.5709 - val_loss: 1.4611 - val_accuracy: 0.5157
Epoch 77/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3609 - accuracy: 0.5709 - val_loss: 1.4334 - val_accuracy: 0.5287
Epoch 78/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3540 - accuracy: 0.5790 - val_loss: 1.4423 - val_accuracy: 0.5213
Epoch 79/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3491 - accuracy: 0.5762 - val_loss: 1.4080 - val_accuracy: 0.5387
Epoch 80/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3497 - accuracy: 0.5795 - val_loss: 1.4128 - val_accuracy: 0.5397
Epoch 81/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3437 - accuracy: 0.5784 - val_loss: 1.4557 - val_accuracy: 0.5060
Epoch 82/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3326 - accuracy: 0.5783 - val_loss: 1.4148 - val_accuracy: 0.5353
Epoch 83/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3280 - accuracy: 0.5833 - val_loss: 1.4037 - val_accuracy: 0.5327
Epoch 84/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3254 - accuracy: 0.5834 - val_loss: 1.5457 - val_accuracy: 0.4803
Epoch 85/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3310 - accuracy: 0.5829 - val_loss: 1.4023 - val_accuracy: 0.5553
Epoch 86/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3216 - accuracy: 0.5864 - val_loss: 1.4342 - val_accuracy: 0.5297
Epoch 87/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3106 - accuracy: 0.5920 - val_loss: 1.3658 - val_accuracy: 0.5560
Epoch 88/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3153 - accuracy: 0.5875 - val_loss: 1.4152 - val_accuracy: 0.5387
Epoch 89/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3065 - accuracy: 0.5886 - val_loss: 1.3658 - val_accuracy: 0.5593
Epoch 90/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3019 - accuracy: 0.5903 - val_loss: 1.4092 - val_accuracy: 0.5297
Epoch 91/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2973 - accuracy: 0.5963 - val_loss: 1.3488 - val_accuracy: 0.5513
Epoch 92/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2925 - accuracy: 0.5953 - val_loss: 1.3673 - val_accuracy: 0.5620
Epoch 93/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2899 - accuracy: 0.5978 - val_loss: 1.4025 - val_accuracy: 0.5320
Epoch 94/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2867 - accuracy: 0.5948 - val_loss: 1.3505 - val_accuracy: 0.5617
Epoch 95/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2843 - accuracy: 0.5963 - val_loss: 1.3670 - val_accuracy: 0.5460
Epoch 96/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2785 - accuracy: 0.5980 - val_loss: 1.3541 - val_accuracy: 0.5537
Epoch 97/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2700 - accuracy: 0.6027 - val_loss: 1.3366 - val_accuracy: 0.5670
Epoch 98/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2747 - accuracy: 0.6011 - val_loss: 1.3356 - val_accuracy: 0.5607
Epoch 99/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2640 - accuracy: 0.6038 - val_loss: 1.3253 - val_accuracy: 0.5630
Epoch 100/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2626 - accuracy: 0.6053 - val_loss: 1.3126 - val_accuracy: 0.5737
Epoch 101/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2594 - accuracy: 0.6054 - val_loss: 1.3368 - val_accuracy: 0.5660
Epoch 102/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2496 - accuracy: 0.6105 - val_loss: 1.3375 - val_accuracy: 0.5623
Epoch 103/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2495 - accuracy: 0.6107 - val_loss: 1.3909 - val_accuracy: 0.5373
Epoch 104/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2440 - accuracy: 0.6071 - val_loss: 1.3363 - val_accuracy: 0.5703
Epoch 105/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2340 - accuracy: 0.6153 - val_loss: 1.2845 - val_accuracy: 0.5797
Epoch 106/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2379 - accuracy: 0.6136 - val_loss: 1.3927 - val_accuracy: 0.5383
Epoch 107/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2392 - accuracy: 0.6066 - val_loss: 1.2860 - val_accuracy: 0.5840
Epoch 108/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2322 - accuracy: 0.6143 - val_loss: 1.2874 - val_accuracy: 0.5883
Epoch 109/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2298 - accuracy: 0.6161 - val_loss: 1.2884 - val_accuracy: 0.5777
Epoch 110/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2222 - accuracy: 0.6170 - val_loss: 1.2650 - val_accuracy: 0.6000
Epoch 111/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2106 - accuracy: 0.6247 - val_loss: 1.2652 - val_accuracy: 0.5950
Epoch 112/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2155 - accuracy: 0.6198 - val_loss: 1.2752 - val_accuracy: 0.5870
Epoch 113/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2141 - accuracy: 0.6180 - val_loss: 1.2787 - val_accuracy: 0.5770
Epoch 114/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2108 - accuracy: 0.6221 - val_loss: 1.2637 - val_accuracy: 0.5997
Epoch 115/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2049 - accuracy: 0.6223 - val_loss: 1.2640 - val_accuracy: 0.5820
Epoch 116/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2102 - accuracy: 0.6217 - val_loss: 1.2800 - val_accuracy: 0.5747
Epoch 117/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2085 - accuracy: 0.6243 - val_loss: 1.3009 - val_accuracy: 0.5680
Epoch 118/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1979 - accuracy: 0.6232 - val_loss: 1.2458 - val_accuracy: 0.5970
Epoch 119/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1958 - accuracy: 0.6241 - val_loss: 1.3023 - val_accuracy: 0.5700
Epoch 120/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1999 - accuracy: 0.6265 - val_loss: 1.2232 - val_accuracy: 0.6100
Epoch 121/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1875 - accuracy: 0.6316 - val_loss: 1.2456 - val_accuracy: 0.5963
Epoch 122/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1891 - accuracy: 0.6299 - val_loss: 1.2306 - val_accuracy: 0.6007
Epoch 123/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1784 - accuracy: 0.6333 - val_loss: 1.2213 - val_accuracy: 0.6090
Epoch 124/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1751 - accuracy: 0.6311 - val_loss: 1.2090 - val_accuracy: 0.6110
Epoch 125/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1758 - accuracy: 0.6306 - val_loss: 1.2977 - val_accuracy: 0.5780
Epoch 126/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1671 - accuracy: 0.6349 - val_loss: 1.2071 - val_accuracy: 0.6110
Epoch 127/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1667 - accuracy: 0.6324 - val_loss: 1.2296 - val_accuracy: 0.5990
Epoch 128/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1616 - accuracy: 0.6374 - val_loss: 1.2524 - val_accuracy: 0.5927
Epoch 129/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1640 - accuracy: 0.6368 - val_loss: 1.2445 - val_accuracy: 0.5973
Epoch 130/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1557 - accuracy: 0.6334 - val_loss: 1.1929 - val_accuracy: 0.6077
Epoch 131/150
283/283 [==============================] - 5s 17ms/step - loss: 1.1567 - accuracy: 0.6336 - val_loss: 1.2147 - val_accuracy: 0.6137
Epoch 132/150
283/283 [==============================] - 5s 17ms/step - loss: 1.1539 - accuracy: 0.6398 - val_loss: 1.3392 - val_accuracy: 0.5600
Epoch 133/150
283/283 [==============================] - 5s 17ms/step - loss: 1.1595 - accuracy: 0.6376 - val_loss: 1.2181 - val_accuracy: 0.6097
Epoch 134/150
283/283 [==============================] - 5s 17ms/step - loss: 1.1444 - accuracy: 0.6420 - val_loss: 1.2162 - val_accuracy: 0.6093
Epoch 135/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1410 - accuracy: 0.6427 - val_loss: 1.1900 - val_accuracy: 0.6063
Epoch 136/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1478 - accuracy: 0.6383 - val_loss: 1.1700 - val_accuracy: 0.6253
Epoch 137/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1338 - accuracy: 0.6443 - val_loss: 1.1792 - val_accuracy: 0.6117
Epoch 138/150
283/283 [==============================] - 5s 17ms/step - loss: 1.1363 - accuracy: 0.6443 - val_loss: 1.1894 - val_accuracy: 0.6140
Epoch 139/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1398 - accuracy: 0.6452 - val_loss: 1.1522 - val_accuracy: 0.6310
Epoch 140/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1262 - accuracy: 0.6478 - val_loss: 1.1833 - val_accuracy: 0.6210
Epoch 141/150
283/283 [==============================] - 5s 17ms/step - loss: 1.1231 - accuracy: 0.6482 - val_loss: 1.3564 - val_accuracy: 0.5613
Epoch 142/150
283/283 [==============================] - 5s 17ms/step - loss: 1.1277 - accuracy: 0.6432 - val_loss: 1.2070 - val_accuracy: 0.6023
Epoch 143/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1258 - accuracy: 0.6421 - val_loss: 1.1453 - val_accuracy: 0.6393
Epoch 144/150
283/283 [==============================] - 5s 17ms/step - loss: 1.1238 - accuracy: 0.6459 - val_loss: 1.1655 - val_accuracy: 0.6227
Epoch 145/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1130 - accuracy: 0.6481 - val_loss: 1.1570 - val_accuracy: 0.6240
Epoch 146/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1153 - accuracy: 0.6525 - val_loss: 1.1670 - val_accuracy: 0.6183
Epoch 147/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1113 - accuracy: 0.6498 - val_loss: 1.1578 - val_accuracy: 0.6290
Epoch 148/150
283/283 [==============================] - 5s 16ms/step - loss: 1.1086 - accuracy: 0.6518 - val_loss: 1.1529 - val_accuracy: 0.6353
Epoch 149/150
283/283 [==============================] - 5s 17ms/step - loss: 1.1076 - accuracy: 0.6540 - val_loss: 1.1377 - val_accuracy: 0.6403
Epoch 150/150
283/283 [==============================] - 5s 17ms/step - loss: 1.1001 - accuracy: 0.6524 - val_loss: 1.1963 - val_accuracy: 0.6060

---

Greyscale validation (128 x 128 image)

---

history = baseline128_model.fit(res128greyscale_train_data, validation_data = res128greyscale_validation_data, epochs = 150, callbacks = earlyStopper)
cnnf.class_image(res128greyscale_full_data, class_names, cmap = "gray")
cnnf.model_metric_graph(history, baseline128_model, res128greyscale_validation_data, name = "baseline model for 128x128 image (normally greyscaled)", class_names = class_names)

Epoch 1/150
283/283 [==============================] - 6s 18ms/step - loss: 2.6408 - accuracy: 0.1078 - val_loss: 2.7554 - val_accuracy: 0.1153
Epoch 2/150
283/283 [==============================] - 5s 18ms/step - loss: 2.5569 - accuracy: 0.1781 - val_loss: 2.6903 - val_accuracy: 0.1213
Epoch 3/150
283/283 [==============================] - 5s 18ms/step - loss: 2.4801 - accuracy: 0.2049 - val_loss: 2.6167 - val_accuracy: 0.1713
Epoch 4/150
283/283 [==============================] - 6s 18ms/step - loss: 2.4137 - accuracy: 0.2424 - val_loss: 2.5369 - val_accuracy: 0.1783
Epoch 5/150
283/283 [==============================] - 6s 20ms/step - loss: 2.3513 - accuracy: 0.2604 - val_loss: 2.5079 - val_accuracy: 0.1960
Epoch 6/150
283/283 [==============================] - 6s 18ms/step - loss: 2.2922 - accuracy: 0.2734 - val_loss: 2.4538 - val_accuracy: 0.2010
Epoch 7/150
283/283 [==============================] - 6s 18ms/step - loss: 2.2458 - accuracy: 0.2854 - val_loss: 2.4360 - val_accuracy: 0.1957
Epoch 8/150
283/283 [==============================] - 6s 19ms/step - loss: 2.2075 - accuracy: 0.2951 - val_loss: 2.4675 - val_accuracy: 0.1993
Epoch 9/150
283/283 [==============================] - 6s 19ms/step - loss: 2.1802 - accuracy: 0.2985 - val_loss: 2.2598 - val_accuracy: 0.2677
Epoch 10/150
283/283 [==============================] - 5s 18ms/step - loss: 2.1506 - accuracy: 0.3106 - val_loss: 2.2515 - val_accuracy: 0.2597
Epoch 11/150
283/283 [==============================] - 6s 18ms/step - loss: 2.1289 - accuracy: 0.3199 - val_loss: 2.2675 - val_accuracy: 0.2517
Epoch 12/150
283/283 [==============================] - 6s 19ms/step - loss: 2.1175 - accuracy: 0.3219 - val_loss: 2.2753 - val_accuracy: 0.2287
Epoch 13/150
283/283 [==============================] - 5s 18ms/step - loss: 2.0991 - accuracy: 0.3242 - val_loss: 2.1941 - val_accuracy: 0.2947
Epoch 14/150
283/283 [==============================] - 5s 17ms/step - loss: 2.0916 - accuracy: 0.3351 - val_loss: 2.1938 - val_accuracy: 0.2753
Epoch 15/150
283/283 [==============================] - 6s 19ms/step - loss: 2.0719 - accuracy: 0.3394 - val_loss: 2.1759 - val_accuracy: 0.2850
Epoch 16/150
283/283 [==============================] - 5s 18ms/step - loss: 2.0613 - accuracy: 0.3454 - val_loss: 2.1800 - val_accuracy: 0.2897
Epoch 17/150
283/283 [==============================] - 5s 18ms/step - loss: 2.0495 - accuracy: 0.3522 - val_loss: 2.1498 - val_accuracy: 0.2917
Epoch 18/150
283/283 [==============================] - 6s 19ms/step - loss: 2.0433 - accuracy: 0.3502 - val_loss: 2.1099 - val_accuracy: 0.3187
Epoch 19/150
283/283 [==============================] - 6s 19ms/step - loss: 2.0276 - accuracy: 0.3595 - val_loss: 2.0965 - val_accuracy: 0.3213
Epoch 20/150
283/283 [==============================] - 6s 19ms/step - loss: 2.0223 - accuracy: 0.3636 - val_loss: 2.0788 - val_accuracy: 0.3393
Epoch 21/150
283/283 [==============================] - 5s 18ms/step - loss: 2.0046 - accuracy: 0.3720 - val_loss: 2.1813 - val_accuracy: 0.2923
Epoch 22/150
283/283 [==============================] - 5s 18ms/step - loss: 1.9947 - accuracy: 0.3761 - val_loss: 2.0978 - val_accuracy: 0.3210
Epoch 23/150
283/283 [==============================] - 5s 18ms/step - loss: 1.9876 - accuracy: 0.3726 - val_loss: 2.1072 - val_accuracy: 0.3157
Epoch 24/150
283/283 [==============================] - 6s 18ms/step - loss: 1.9760 - accuracy: 0.3881 - val_loss: 2.0860 - val_accuracy: 0.3280
Epoch 25/150
283/283 [==============================] - 6s 19ms/step - loss: 1.9658 - accuracy: 0.3875 - val_loss: 2.0313 - val_accuracy: 0.3503
Epoch 26/150
283/283 [==============================] - 5s 18ms/step - loss: 1.9505 - accuracy: 0.3953 - val_loss: 2.0213 - val_accuracy: 0.3583
Epoch 27/150
283/283 [==============================] - 5s 18ms/step - loss: 1.9359 - accuracy: 0.4000 - val_loss: 2.0214 - val_accuracy: 0.3450
Epoch 28/150
283/283 [==============================] - 5s 17ms/step - loss: 1.9220 - accuracy: 0.3996 - val_loss: 1.9815 - val_accuracy: 0.3650
Epoch 29/150
283/283 [==============================] - 5s 18ms/step - loss: 1.9169 - accuracy: 0.4023 - val_loss: 1.9588 - val_accuracy: 0.3870
Epoch 30/150
283/283 [==============================] - 5s 17ms/step - loss: 1.9066 - accuracy: 0.4068 - val_loss: 1.9691 - val_accuracy: 0.3713
Epoch 31/150
283/283 [==============================] - 6s 19ms/step - loss: 1.8877 - accuracy: 0.4138 - val_loss: 1.9966 - val_accuracy: 0.3553
Epoch 32/150
283/283 [==============================] - 6s 19ms/step - loss: 1.8709 - accuracy: 0.4276 - val_loss: 1.9611 - val_accuracy: 0.3627
Epoch 33/150
283/283 [==============================] - 6s 18ms/step - loss: 1.8757 - accuracy: 0.4202 - val_loss: 1.9387 - val_accuracy: 0.3720
Epoch 34/150
283/283 [==============================] - 6s 18ms/step - loss: 1.8519 - accuracy: 0.4322 - val_loss: 1.9375 - val_accuracy: 0.3963
Epoch 35/150
283/283 [==============================] - 6s 19ms/step - loss: 1.8411 - accuracy: 0.4332 - val_loss: 1.9361 - val_accuracy: 0.3873
Epoch 36/150
283/283 [==============================] - 6s 19ms/step - loss: 1.8367 - accuracy: 0.4384 - val_loss: 1.9093 - val_accuracy: 0.3997
Epoch 37/150
283/283 [==============================] - 5s 18ms/step - loss: 1.8243 - accuracy: 0.4405 - val_loss: 1.9763 - val_accuracy: 0.3630
Epoch 38/150
283/283 [==============================] - 5s 17ms/step - loss: 1.8135 - accuracy: 0.4463 - val_loss: 1.9547 - val_accuracy: 0.3853
Epoch 39/150
283/283 [==============================] - 5s 18ms/step - loss: 1.8038 - accuracy: 0.4486 - val_loss: 1.9395 - val_accuracy: 0.3920
Epoch 40/150
283/283 [==============================] - 5s 18ms/step - loss: 1.7931 - accuracy: 0.4540 - val_loss: 1.8492 - val_accuracy: 0.4483
Epoch 41/150
283/283 [==============================] - 5s 18ms/step - loss: 1.7885 - accuracy: 0.4520 - val_loss: 1.8518 - val_accuracy: 0.4503
Epoch 42/150
283/283 [==============================] - 5s 18ms/step - loss: 1.7734 - accuracy: 0.4611 - val_loss: 1.8888 - val_accuracy: 0.4187
Epoch 43/150
283/283 [==============================] - 5s 18ms/step - loss: 1.7679 - accuracy: 0.4658 - val_loss: 1.8294 - val_accuracy: 0.4370
Epoch 44/150
283/283 [==============================] - 5s 17ms/step - loss: 1.7605 - accuracy: 0.4656 - val_loss: 1.8837 - val_accuracy: 0.4140
Epoch 45/150
283/283 [==============================] - 5s 18ms/step - loss: 1.7411 - accuracy: 0.4751 - val_loss: 1.8037 - val_accuracy: 0.4680
Epoch 46/150
283/283 [==============================] - 5s 18ms/step - loss: 1.7360 - accuracy: 0.4771 - val_loss: 1.8091 - val_accuracy: 0.4377
Epoch 47/150
283/283 [==============================] - 5s 18ms/step - loss: 1.7312 - accuracy: 0.4765 - val_loss: 1.7957 - val_accuracy: 0.4687
Epoch 48/150
283/283 [==============================] - 6s 18ms/step - loss: 1.7172 - accuracy: 0.4833 - val_loss: 1.7650 - val_accuracy: 0.4840
Epoch 49/150
283/283 [==============================] - 5s 18ms/step - loss: 1.7136 - accuracy: 0.4858 - val_loss: 1.8339 - val_accuracy: 0.4307
Epoch 50/150
283/283 [==============================] - 5s 18ms/step - loss: 1.7037 - accuracy: 0.4863 - val_loss: 1.7736 - val_accuracy: 0.4537
Epoch 51/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6956 - accuracy: 0.4920 - val_loss: 1.8181 - val_accuracy: 0.4503
Epoch 52/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6832 - accuracy: 0.4916 - val_loss: 1.8062 - val_accuracy: 0.4573
Epoch 53/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6740 - accuracy: 0.5009 - val_loss: 1.7758 - val_accuracy: 0.4727
Epoch 54/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6680 - accuracy: 0.4984 - val_loss: 1.7388 - val_accuracy: 0.4780
Epoch 55/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6595 - accuracy: 0.4998 - val_loss: 1.7511 - val_accuracy: 0.4533
Epoch 56/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6494 - accuracy: 0.5045 - val_loss: 1.7124 - val_accuracy: 0.4857
Epoch 57/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6423 - accuracy: 0.5089 - val_loss: 1.7220 - val_accuracy: 0.4927
Epoch 58/150
283/283 [==============================] - 6s 19ms/step - loss: 1.6353 - accuracy: 0.5104 - val_loss: 1.6820 - val_accuracy: 0.4983
Epoch 59/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6273 - accuracy: 0.5134 - val_loss: 1.6962 - val_accuracy: 0.4887
Epoch 60/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6208 - accuracy: 0.5160 - val_loss: 1.7033 - val_accuracy: 0.4930
Epoch 61/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6095 - accuracy: 0.5148 - val_loss: 1.6676 - val_accuracy: 0.5087
Epoch 62/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6030 - accuracy: 0.5294 - val_loss: 1.7004 - val_accuracy: 0.4897
Epoch 63/150
283/283 [==============================] - 5s 17ms/step - loss: 1.6032 - accuracy: 0.5257 - val_loss: 1.6837 - val_accuracy: 0.4997
Epoch 64/150
283/283 [==============================] - 5s 17ms/step - loss: 1.5950 - accuracy: 0.5286 - val_loss: 1.7179 - val_accuracy: 0.4693
Epoch 65/150
283/283 [==============================] - 5s 18ms/step - loss: 1.5905 - accuracy: 0.5243 - val_loss: 1.6618 - val_accuracy: 0.5150
Epoch 66/150
283/283 [==============================] - 5s 18ms/step - loss: 1.5827 - accuracy: 0.5270 - val_loss: 1.6794 - val_accuracy: 0.5080
Epoch 67/150
283/283 [==============================] - 6s 20ms/step - loss: 1.5781 - accuracy: 0.5274 - val_loss: 1.6159 - val_accuracy: 0.5247
Epoch 68/150
283/283 [==============================] - 6s 19ms/step - loss: 1.5667 - accuracy: 0.5309 - val_loss: 1.6944 - val_accuracy: 0.4843
Epoch 69/150
283/283 [==============================] - 6s 21ms/step - loss: 1.5549 - accuracy: 0.5370 - val_loss: 1.6174 - val_accuracy: 0.5193
Epoch 70/150
283/283 [==============================] - 6s 20ms/step - loss: 1.5581 - accuracy: 0.5359 - val_loss: 1.6127 - val_accuracy: 0.5260
Epoch 71/150
283/283 [==============================] - 6s 20ms/step - loss: 1.5469 - accuracy: 0.5422 - val_loss: 1.6820 - val_accuracy: 0.4877
Epoch 72/150
283/283 [==============================] - 6s 20ms/step - loss: 1.5436 - accuracy: 0.5445 - val_loss: 1.6410 - val_accuracy: 0.5137
Epoch 73/150
283/283 [==============================] - 6s 18ms/step - loss: 1.5414 - accuracy: 0.5407 - val_loss: 1.6027 - val_accuracy: 0.5353
Epoch 74/150
283/283 [==============================] - 5s 17ms/step - loss: 1.5367 - accuracy: 0.5434 - val_loss: 1.6067 - val_accuracy: 0.5283
Epoch 75/150
283/283 [==============================] - 5s 17ms/step - loss: 1.5281 - accuracy: 0.5423 - val_loss: 1.6204 - val_accuracy: 0.5190
Epoch 76/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5235 - accuracy: 0.5449 - val_loss: 1.7209 - val_accuracy: 0.4730
Epoch 77/150
283/283 [==============================] - 5s 16ms/step - loss: 1.5173 - accuracy: 0.5477 - val_loss: 1.6750 - val_accuracy: 0.4800
Epoch 78/150
283/283 [==============================] - 5s 17ms/step - loss: 1.5107 - accuracy: 0.5471 - val_loss: 1.5528 - val_accuracy: 0.5357
Epoch 79/150
283/283 [==============================] - 5s 17ms/step - loss: 1.5113 - accuracy: 0.5487 - val_loss: 1.5890 - val_accuracy: 0.5280
Epoch 80/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4994 - accuracy: 0.5496 - val_loss: 1.5359 - val_accuracy: 0.5483
Epoch 81/150
283/283 [==============================] - 5s 17ms/step - loss: 1.5042 - accuracy: 0.5512 - val_loss: 1.5724 - val_accuracy: 0.5277
Epoch 82/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4933 - accuracy: 0.5527 - val_loss: 1.6636 - val_accuracy: 0.4853
Epoch 83/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4902 - accuracy: 0.5562 - val_loss: 1.5724 - val_accuracy: 0.5290
Epoch 84/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4823 - accuracy: 0.5570 - val_loss: 1.5807 - val_accuracy: 0.5267
Epoch 85/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4789 - accuracy: 0.5577 - val_loss: 1.6333 - val_accuracy: 0.5057
Epoch 86/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4813 - accuracy: 0.5567 - val_loss: 1.6051 - val_accuracy: 0.5147
Epoch 87/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4701 - accuracy: 0.5616 - val_loss: 1.5663 - val_accuracy: 0.5360
Epoch 88/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4765 - accuracy: 0.5555 - val_loss: 1.5226 - val_accuracy: 0.5450
Epoch 89/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4666 - accuracy: 0.5573 - val_loss: 1.5681 - val_accuracy: 0.5257
Epoch 90/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4581 - accuracy: 0.5604 - val_loss: 1.5216 - val_accuracy: 0.5390
Epoch 91/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4554 - accuracy: 0.5606 - val_loss: 1.5369 - val_accuracy: 0.5327
Epoch 92/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4616 - accuracy: 0.5601 - val_loss: 1.5054 - val_accuracy: 0.5630
Epoch 93/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4515 - accuracy: 0.5654 - val_loss: 1.5471 - val_accuracy: 0.5260
Epoch 94/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4472 - accuracy: 0.5666 - val_loss: 1.5122 - val_accuracy: 0.5550
Epoch 95/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4445 - accuracy: 0.5632 - val_loss: 1.6064 - val_accuracy: 0.5150
Epoch 96/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4446 - accuracy: 0.5672 - val_loss: 1.5373 - val_accuracy: 0.5357
Epoch 97/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4387 - accuracy: 0.5638 - val_loss: 1.5082 - val_accuracy: 0.5483
Epoch 98/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4301 - accuracy: 0.5724 - val_loss: 1.5081 - val_accuracy: 0.5513
Epoch 99/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4323 - accuracy: 0.5654 - val_loss: 1.5292 - val_accuracy: 0.5363
Epoch 100/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4253 - accuracy: 0.5680 - val_loss: 1.5145 - val_accuracy: 0.5387
Epoch 101/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4227 - accuracy: 0.5740 - val_loss: 1.4716 - val_accuracy: 0.5467
Epoch 102/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4163 - accuracy: 0.5722 - val_loss: 1.4802 - val_accuracy: 0.5573
Epoch 103/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4150 - accuracy: 0.5701 - val_loss: 1.4681 - val_accuracy: 0.5537
Epoch 104/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4183 - accuracy: 0.5760 - val_loss: 1.4959 - val_accuracy: 0.5310
Epoch 105/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4163 - accuracy: 0.5743 - val_loss: 1.5400 - val_accuracy: 0.5423
Epoch 106/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4133 - accuracy: 0.5735 - val_loss: 1.4785 - val_accuracy: 0.5500
Epoch 107/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4026 - accuracy: 0.5740 - val_loss: 1.4429 - val_accuracy: 0.5727
Epoch 108/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4078 - accuracy: 0.5709 - val_loss: 1.4698 - val_accuracy: 0.5563
Epoch 109/150
283/283 [==============================] - 5s 16ms/step - loss: 1.4071 - accuracy: 0.5737 - val_loss: 1.4532 - val_accuracy: 0.5620
Epoch 110/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3934 - accuracy: 0.5805 - val_loss: 1.4662 - val_accuracy: 0.5503
Epoch 111/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3898 - accuracy: 0.5842 - val_loss: 1.4966 - val_accuracy: 0.5387
Epoch 112/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3862 - accuracy: 0.5807 - val_loss: 1.4551 - val_accuracy: 0.5667
Epoch 113/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3894 - accuracy: 0.5810 - val_loss: 1.4884 - val_accuracy: 0.5450
Epoch 114/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3911 - accuracy: 0.5780 - val_loss: 1.4122 - val_accuracy: 0.5780
Epoch 115/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3874 - accuracy: 0.5778 - val_loss: 1.4835 - val_accuracy: 0.5647
Epoch 116/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3852 - accuracy: 0.5780 - val_loss: 1.5265 - val_accuracy: 0.5337
Epoch 117/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3811 - accuracy: 0.5823 - val_loss: 1.4247 - val_accuracy: 0.5653
Epoch 118/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3750 - accuracy: 0.5873 - val_loss: 1.4476 - val_accuracy: 0.5693
Epoch 119/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3636 - accuracy: 0.5848 - val_loss: 1.4334 - val_accuracy: 0.5603
Epoch 120/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3708 - accuracy: 0.5862 - val_loss: 1.4215 - val_accuracy: 0.5763
Epoch 121/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3659 - accuracy: 0.5850 - val_loss: 1.4714 - val_accuracy: 0.5637
Epoch 122/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3658 - accuracy: 0.5848 - val_loss: 1.3973 - val_accuracy: 0.5843
Epoch 123/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3586 - accuracy: 0.5865 - val_loss: 1.5174 - val_accuracy: 0.5393
Epoch 124/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3564 - accuracy: 0.5876 - val_loss: 1.5561 - val_accuracy: 0.5093
Epoch 125/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3507 - accuracy: 0.5908 - val_loss: 1.4854 - val_accuracy: 0.5457
Epoch 126/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3561 - accuracy: 0.5846 - val_loss: 1.4535 - val_accuracy: 0.5457
Epoch 127/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3501 - accuracy: 0.5905 - val_loss: 1.4564 - val_accuracy: 0.5510
Epoch 128/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3544 - accuracy: 0.5904 - val_loss: 1.4507 - val_accuracy: 0.5640
Epoch 129/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3478 - accuracy: 0.5955 - val_loss: 1.4210 - val_accuracy: 0.5647
Epoch 130/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3449 - accuracy: 0.5919 - val_loss: 1.4228 - val_accuracy: 0.5617
Epoch 131/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3384 - accuracy: 0.5920 - val_loss: 1.3924 - val_accuracy: 0.5697
Epoch 132/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3481 - accuracy: 0.5897 - val_loss: 1.3883 - val_accuracy: 0.5863
Epoch 133/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3401 - accuracy: 0.5930 - val_loss: 1.4992 - val_accuracy: 0.5330
Epoch 134/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3326 - accuracy: 0.5935 - val_loss: 1.4985 - val_accuracy: 0.5307
Epoch 135/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3294 - accuracy: 0.5925 - val_loss: 1.4420 - val_accuracy: 0.5633
Epoch 136/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3315 - accuracy: 0.5883 - val_loss: 1.3997 - val_accuracy: 0.5750
Epoch 137/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3298 - accuracy: 0.5980 - val_loss: 1.4304 - val_accuracy: 0.5723
Epoch 138/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3297 - accuracy: 0.5943 - val_loss: 1.3814 - val_accuracy: 0.5883
Epoch 139/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3275 - accuracy: 0.5967 - val_loss: 1.3632 - val_accuracy: 0.5757
Epoch 140/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3269 - accuracy: 0.5978 - val_loss: 1.3908 - val_accuracy: 0.5740
Epoch 141/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3182 - accuracy: 0.5981 - val_loss: 1.3574 - val_accuracy: 0.5903
Epoch 142/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3153 - accuracy: 0.5985 - val_loss: 1.4173 - val_accuracy: 0.5690
Epoch 143/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3109 - accuracy: 0.5978 - val_loss: 1.3619 - val_accuracy: 0.5797
Epoch 144/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3101 - accuracy: 0.5998 - val_loss: 1.4027 - val_accuracy: 0.5720
Epoch 145/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3091 - accuracy: 0.5998 - val_loss: 1.3337 - val_accuracy: 0.5943
Epoch 146/150
283/283 [==============================] - 5s 16ms/step - loss: 1.3086 - accuracy: 0.5986 - val_loss: 1.3357 - val_accuracy: 0.5920
Epoch 147/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3051 - accuracy: 0.6018 - val_loss: 1.4908 - val_accuracy: 0.5477
Epoch 148/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3036 - accuracy: 0.6009 - val_loss: 1.3150 - val_accuracy: 0.6073
Epoch 149/150
283/283 [==============================] - 5s 16ms/step - loss: 1.2990 - accuracy: 0.6079 - val_loss: 1.4468 - val_accuracy: 0.5387
Epoch 150/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3008 - accuracy: 0.6058 - val_loss: 1.3631 - val_accuracy: 0.5850

baseline128greyscale_model = clone_model(baseline_model)
baseline128greyscale_model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])

history = baseline128luminosity_model.fit(res128luminosity_train_data, validation_data = res128luminosity_validation_data, epochs = 150, callbacks = earlyStopper)
cnnf.class_image(res128luminosity_full_data, class_names, cmap = "gray")
cnnf.model_metric_graph(history, baseline128luminosity_model, res128luminosity_validation_data, name = "baseline model for 128x128 image (using luminosity method)", class_names = class_names)

Epoch 1/150
283/283 [==============================] - 5s 17ms/step - loss: 2.6276 - accuracy: 0.1135 - val_loss: 2.7736 - val_accuracy: 0.1150
Epoch 2/150
283/283 [==============================] - 5s 17ms/step - loss: 2.5313 - accuracy: 0.1917 - val_loss: 2.6425 - val_accuracy: 0.1773
Epoch 3/150
283/283 [==============================] - 5s 17ms/step - loss: 2.4202 - accuracy: 0.2336 - val_loss: 2.5386 - val_accuracy: 0.1793
Epoch 4/150
283/283 [==============================] - 5s 17ms/step - loss: 2.3292 - accuracy: 0.2621 - val_loss: 2.4175 - val_accuracy: 0.2170
Epoch 5/150
283/283 [==============================] - 5s 18ms/step - loss: 2.2522 - accuracy: 0.2785 - val_loss: 2.3220 - val_accuracy: 0.2300
Epoch 6/150
283/283 [==============================] - 5s 17ms/step - loss: 2.2048 - accuracy: 0.2894 - val_loss: 2.2824 - val_accuracy: 0.2573
Epoch 7/150
283/283 [==============================] - 5s 18ms/step - loss: 2.1712 - accuracy: 0.3044 - val_loss: 2.3325 - val_accuracy: 0.2110
Epoch 8/150
283/283 [==============================] - 5s 18ms/step - loss: 2.1456 - accuracy: 0.3137 - val_loss: 2.2341 - val_accuracy: 0.2587
Epoch 9/150
283/283 [==============================] - 5s 17ms/step - loss: 2.1242 - accuracy: 0.3206 - val_loss: 2.2262 - val_accuracy: 0.2677
Epoch 10/150
283/283 [==============================] - 5s 18ms/step - loss: 2.1110 - accuracy: 0.3270 - val_loss: 2.1865 - val_accuracy: 0.2830
Epoch 11/150
283/283 [==============================] - 5s 18ms/step - loss: 2.0892 - accuracy: 0.3311 - val_loss: 2.1513 - val_accuracy: 0.3097
Epoch 12/150
283/283 [==============================] - 5s 18ms/step - loss: 2.0762 - accuracy: 0.3386 - val_loss: 2.1308 - val_accuracy: 0.3183
Epoch 13/150
283/283 [==============================] - 5s 17ms/step - loss: 2.0585 - accuracy: 0.3487 - val_loss: 2.1853 - val_accuracy: 0.2903
Epoch 14/150
283/283 [==============================] - 5s 18ms/step - loss: 2.0493 - accuracy: 0.3539 - val_loss: 2.1824 - val_accuracy: 0.2803
Epoch 15/150
283/283 [==============================] - 5s 18ms/step - loss: 2.0341 - accuracy: 0.3576 - val_loss: 2.1780 - val_accuracy: 0.2913
Epoch 16/150
283/283 [==============================] - 5s 18ms/step - loss: 2.0240 - accuracy: 0.3614 - val_loss: 2.1852 - val_accuracy: 0.2783
Epoch 17/150
283/283 [==============================] - 5s 18ms/step - loss: 2.0026 - accuracy: 0.3706 - val_loss: 2.0520 - val_accuracy: 0.3560
Epoch 18/150
283/283 [==============================] - 5s 17ms/step - loss: 1.9892 - accuracy: 0.3741 - val_loss: 2.1072 - val_accuracy: 0.3090
Epoch 19/150
283/283 [==============================] - 5s 18ms/step - loss: 1.9737 - accuracy: 0.3789 - val_loss: 2.0387 - val_accuracy: 0.3503
Epoch 20/150
283/283 [==============================] - 6s 18ms/step - loss: 1.9636 - accuracy: 0.3816 - val_loss: 2.0070 - val_accuracy: 0.3780
Epoch 21/150
283/283 [==============================] - 5s 18ms/step - loss: 1.9433 - accuracy: 0.3914 - val_loss: 1.9794 - val_accuracy: 0.3897
Epoch 22/150
283/283 [==============================] - 5s 18ms/step - loss: 1.9280 - accuracy: 0.3939 - val_loss: 2.0382 - val_accuracy: 0.3553
Epoch 23/150
283/283 [==============================] - 5s 18ms/step - loss: 1.9141 - accuracy: 0.4041 - val_loss: 1.9950 - val_accuracy: 0.3673
Epoch 24/150
283/283 [==============================] - 6s 19ms/step - loss: 1.8903 - accuracy: 0.4107 - val_loss: 1.9797 - val_accuracy: 0.3850
Epoch 25/150
283/283 [==============================] - 6s 18ms/step - loss: 1.8816 - accuracy: 0.4071 - val_loss: 2.0751 - val_accuracy: 0.3353
Epoch 26/150
283/283 [==============================] - 6s 19ms/step - loss: 1.8627 - accuracy: 0.4177 - val_loss: 1.9136 - val_accuracy: 0.4127
Epoch 27/150
283/283 [==============================] - 6s 18ms/step - loss: 1.8436 - accuracy: 0.4289 - val_loss: 1.9300 - val_accuracy: 0.3997
Epoch 28/150
283/283 [==============================] - 5s 18ms/step - loss: 1.8327 - accuracy: 0.4387 - val_loss: 1.9118 - val_accuracy: 0.4077
Epoch 29/150
283/283 [==============================] - 6s 18ms/step - loss: 1.8131 - accuracy: 0.4426 - val_loss: 1.8724 - val_accuracy: 0.4420
Epoch 30/150
283/283 [==============================] - 5s 18ms/step - loss: 1.7999 - accuracy: 0.4500 - val_loss: 1.8532 - val_accuracy: 0.4473
Epoch 31/150
283/283 [==============================] - 6s 18ms/step - loss: 1.7837 - accuracy: 0.4536 - val_loss: 1.8855 - val_accuracy: 0.4160
Epoch 32/150
283/283 [==============================] - 5s 18ms/step - loss: 1.7685 - accuracy: 0.4584 - val_loss: 1.8719 - val_accuracy: 0.4043
Epoch 33/150
283/283 [==============================] - 5s 17ms/step - loss: 1.7571 - accuracy: 0.4652 - val_loss: 1.9498 - val_accuracy: 0.3860
Epoch 34/150
283/283 [==============================] - 5s 18ms/step - loss: 1.7378 - accuracy: 0.4708 - val_loss: 1.8155 - val_accuracy: 0.4523
Epoch 35/150
283/283 [==============================] - 6s 21ms/step - loss: 1.7361 - accuracy: 0.4677 - val_loss: 1.7904 - val_accuracy: 0.4587
Epoch 36/150
283/283 [==============================] - 6s 19ms/step - loss: 1.7173 - accuracy: 0.4788 - val_loss: 1.7946 - val_accuracy: 0.4577
Epoch 37/150
283/283 [==============================] - 6s 19ms/step - loss: 1.7122 - accuracy: 0.4833 - val_loss: 1.8564 - val_accuracy: 0.4253
Epoch 38/150
283/283 [==============================] - 6s 19ms/step - loss: 1.7061 - accuracy: 0.4842 - val_loss: 1.7850 - val_accuracy: 0.4727
Epoch 39/150
283/283 [==============================] - 6s 18ms/step - loss: 1.6926 - accuracy: 0.4909 - val_loss: 1.7868 - val_accuracy: 0.4580
Epoch 40/150
283/283 [==============================] - 5s 18ms/step - loss: 1.6857 - accuracy: 0.4942 - val_loss: 1.7460 - val_accuracy: 0.4823
Epoch 41/150
283/283 [==============================] - 6s 19ms/step - loss: 1.6729 - accuracy: 0.4940 - val_loss: 1.7757 - val_accuracy: 0.4693
Epoch 42/150
283/283 [==============================] - 6s 19ms/step - loss: 1.6597 - accuracy: 0.5019 - val_loss: 1.7433 - val_accuracy: 0.4717
Epoch 43/150
283/283 [==============================] - 6s 19ms/step - loss: 1.6536 - accuracy: 0.5020 - val_loss: 1.7035 - val_accuracy: 0.4857
Epoch 44/150
283/283 [==============================] - 6s 18ms/step - loss: 1.6484 - accuracy: 0.5039 - val_loss: 1.7168 - val_accuracy: 0.5000
Epoch 45/150
283/283 [==============================] - 6s 19ms/step - loss: 1.6363 - accuracy: 0.5113 - val_loss: 1.7126 - val_accuracy: 0.4830
Epoch 46/150
283/283 [==============================] - 5s 18ms/step - loss: 1.6333 - accuracy: 0.5141 - val_loss: 1.6959 - val_accuracy: 0.4870
Epoch 47/150
283/283 [==============================] - 5s 18ms/step - loss: 1.6327 - accuracy: 0.5091 - val_loss: 1.7462 - val_accuracy: 0.4737
Epoch 48/150
283/283 [==============================] - 5s 18ms/step - loss: 1.6158 - accuracy: 0.5168 - val_loss: 1.7156 - val_accuracy: 0.4853
Epoch 49/150
283/283 [==============================] - 6s 19ms/step - loss: 1.6082 - accuracy: 0.5172 - val_loss: 1.6732 - val_accuracy: 0.5067
Epoch 50/150
283/283 [==============================] - 6s 18ms/step - loss: 1.6031 - accuracy: 0.5230 - val_loss: 1.7524 - val_accuracy: 0.4600
Epoch 51/150
283/283 [==============================] - 6s 19ms/step - loss: 1.5923 - accuracy: 0.5246 - val_loss: 1.7314 - val_accuracy: 0.4790
Epoch 52/150
283/283 [==============================] - 6s 18ms/step - loss: 1.5878 - accuracy: 0.5281 - val_loss: 1.6394 - val_accuracy: 0.4953
Epoch 53/150
283/283 [==============================] - 6s 18ms/step - loss: 1.5822 - accuracy: 0.5302 - val_loss: 1.6910 - val_accuracy: 0.5067
Epoch 54/150
283/283 [==============================] - 5s 18ms/step - loss: 1.5841 - accuracy: 0.5219 - val_loss: 1.6668 - val_accuracy: 0.4977
Epoch 55/150
283/283 [==============================] - 6s 18ms/step - loss: 1.5745 - accuracy: 0.5325 - val_loss: 1.6421 - val_accuracy: 0.5227
Epoch 56/150
283/283 [==============================] - 5s 18ms/step - loss: 1.5603 - accuracy: 0.5353 - val_loss: 1.6487 - val_accuracy: 0.5067
Epoch 57/150
283/283 [==============================] - 6s 18ms/step - loss: 1.5534 - accuracy: 0.5382 - val_loss: 1.6352 - val_accuracy: 0.5130
Epoch 58/150
283/283 [==============================] - 5s 18ms/step - loss: 1.5609 - accuracy: 0.5302 - val_loss: 1.6134 - val_accuracy: 0.5197
Epoch 59/150
283/283 [==============================] - 6s 19ms/step - loss: 1.5542 - accuracy: 0.5392 - val_loss: 1.6186 - val_accuracy: 0.5217
Epoch 60/150
283/283 [==============================] - 6s 18ms/step - loss: 1.5400 - accuracy: 0.5444 - val_loss: 1.6576 - val_accuracy: 0.5033
Epoch 61/150
283/283 [==============================] - 6s 18ms/step - loss: 1.5406 - accuracy: 0.5440 - val_loss: 1.6908 - val_accuracy: 0.4820
Epoch 62/150
283/283 [==============================] - 6s 19ms/step - loss: 1.5430 - accuracy: 0.5347 - val_loss: 1.6106 - val_accuracy: 0.5143
Epoch 63/150
283/283 [==============================] - 6s 18ms/step - loss: 1.5331 - accuracy: 0.5425 - val_loss: 1.6243 - val_accuracy: 0.5223
Epoch 64/150
283/283 [==============================] - 5s 18ms/step - loss: 1.5157 - accuracy: 0.5533 - val_loss: 1.5903 - val_accuracy: 0.5253
Epoch 65/150
283/283 [==============================] - 6s 18ms/step - loss: 1.5131 - accuracy: 0.5482 - val_loss: 1.5728 - val_accuracy: 0.5400
Epoch 66/150
283/283 [==============================] - 6s 18ms/step - loss: 1.5220 - accuracy: 0.5444 - val_loss: 1.6151 - val_accuracy: 0.5160
Epoch 67/150
283/283 [==============================] - 6s 18ms/step - loss: 1.5066 - accuracy: 0.5519 - val_loss: 1.5992 - val_accuracy: 0.5190
Epoch 68/150
283/283 [==============================] - 6s 18ms/step - loss: 1.5024 - accuracy: 0.5536 - val_loss: 1.6315 - val_accuracy: 0.4993
Epoch 69/150
283/283 [==============================] - 5s 18ms/step - loss: 1.4999 - accuracy: 0.5538 - val_loss: 1.5635 - val_accuracy: 0.5383
Epoch 70/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4925 - accuracy: 0.5527 - val_loss: 1.5275 - val_accuracy: 0.5543
Epoch 71/150
283/283 [==============================] - 6s 19ms/step - loss: 1.4889 - accuracy: 0.5543 - val_loss: 1.6014 - val_accuracy: 0.5293
Epoch 72/150
283/283 [==============================] - 6s 18ms/step - loss: 1.4865 - accuracy: 0.5572 - val_loss: 1.5454 - val_accuracy: 0.5430
Epoch 73/150
283/283 [==============================] - 5s 18ms/step - loss: 1.4771 - accuracy: 0.5574 - val_loss: 1.5375 - val_accuracy: 0.5370
Epoch 74/150
283/283 [==============================] - 6s 18ms/step - loss: 1.4859 - accuracy: 0.5531 - val_loss: 1.5618 - val_accuracy: 0.5170
Epoch 75/150
283/283 [==============================] - 6s 18ms/step - loss: 1.4725 - accuracy: 0.5639 - val_loss: 1.6893 - val_accuracy: 0.4910
Epoch 76/150
283/283 [==============================] - 6s 19ms/step - loss: 1.4619 - accuracy: 0.5645 - val_loss: 1.5204 - val_accuracy: 0.5510
Epoch 77/150
283/283 [==============================] - 6s 18ms/step - loss: 1.4580 - accuracy: 0.5613 - val_loss: 1.5291 - val_accuracy: 0.5523
Epoch 78/150
283/283 [==============================] - 5s 18ms/step - loss: 1.4620 - accuracy: 0.5647 - val_loss: 1.6186 - val_accuracy: 0.4887
Epoch 79/150
283/283 [==============================] - 5s 18ms/step - loss: 1.4569 - accuracy: 0.5656 - val_loss: 1.4905 - val_accuracy: 0.5590
Epoch 80/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4514 - accuracy: 0.5645 - val_loss: 1.4985 - val_accuracy: 0.5617
Epoch 81/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4461 - accuracy: 0.5667 - val_loss: 1.5102 - val_accuracy: 0.5467
Epoch 82/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4400 - accuracy: 0.5665 - val_loss: 1.5524 - val_accuracy: 0.5193
Epoch 83/150
283/283 [==============================] - 6s 18ms/step - loss: 1.4379 - accuracy: 0.5683 - val_loss: 1.4835 - val_accuracy: 0.5540
Epoch 84/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4385 - accuracy: 0.5693 - val_loss: 1.5261 - val_accuracy: 0.5273
Epoch 85/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4335 - accuracy: 0.5707 - val_loss: 1.5392 - val_accuracy: 0.5443
Epoch 86/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4275 - accuracy: 0.5726 - val_loss: 1.5308 - val_accuracy: 0.5333
Epoch 87/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4221 - accuracy: 0.5748 - val_loss: 1.4860 - val_accuracy: 0.5483
Epoch 88/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4182 - accuracy: 0.5757 - val_loss: 1.4827 - val_accuracy: 0.5597
Epoch 89/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4147 - accuracy: 0.5755 - val_loss: 1.5378 - val_accuracy: 0.5220
Epoch 90/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4070 - accuracy: 0.5821 - val_loss: 1.6334 - val_accuracy: 0.4970
Epoch 91/150
283/283 [==============================] - 5s 18ms/step - loss: 1.4165 - accuracy: 0.5710 - val_loss: 1.4493 - val_accuracy: 0.5670
Epoch 92/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4035 - accuracy: 0.5785 - val_loss: 1.4521 - val_accuracy: 0.5723
Epoch 93/150
283/283 [==============================] - 5s 18ms/step - loss: 1.3995 - accuracy: 0.5845 - val_loss: 1.4441 - val_accuracy: 0.5660
Epoch 94/150
283/283 [==============================] - 5s 17ms/step - loss: 1.4114 - accuracy: 0.5741 - val_loss: 1.4682 - val_accuracy: 0.5610
Epoch 95/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3956 - accuracy: 0.5786 - val_loss: 1.5790 - val_accuracy: 0.5043
Epoch 96/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3920 - accuracy: 0.5803 - val_loss: 1.4451 - val_accuracy: 0.5633
Epoch 97/150
283/283 [==============================] - 6s 18ms/step - loss: 1.3901 - accuracy: 0.5853 - val_loss: 1.4570 - val_accuracy: 0.5497
Epoch 98/150
283/283 [==============================] - 6s 19ms/step - loss: 1.3798 - accuracy: 0.5860 - val_loss: 1.4403 - val_accuracy: 0.5870
Epoch 99/150
283/283 [==============================] - 5s 18ms/step - loss: 1.3862 - accuracy: 0.5799 - val_loss: 1.4347 - val_accuracy: 0.5777
Epoch 100/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3854 - accuracy: 0.5836 - val_loss: 1.4068 - val_accuracy: 0.5890
Epoch 101/150
283/283 [==============================] - 5s 18ms/step - loss: 1.3752 - accuracy: 0.5873 - val_loss: 1.4457 - val_accuracy: 0.5690
Epoch 102/150
283/283 [==============================] - 6s 18ms/step - loss: 1.3673 - accuracy: 0.5862 - val_loss: 1.4547 - val_accuracy: 0.5623
Epoch 103/150
283/283 [==============================] - 5s 18ms/step - loss: 1.3739 - accuracy: 0.5872 - val_loss: 1.4405 - val_accuracy: 0.5880
Epoch 104/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3679 - accuracy: 0.5861 - val_loss: 1.4122 - val_accuracy: 0.5747
Epoch 105/150
283/283 [==============================] - 5s 18ms/step - loss: 1.3569 - accuracy: 0.5905 - val_loss: 1.4242 - val_accuracy: 0.5540
Epoch 106/150
283/283 [==============================] - 5s 18ms/step - loss: 1.3539 - accuracy: 0.5919 - val_loss: 1.4628 - val_accuracy: 0.5467
Epoch 107/150
283/283 [==============================] - 6s 18ms/step - loss: 1.3706 - accuracy: 0.5820 - val_loss: 1.4621 - val_accuracy: 0.5620
Epoch 108/150
283/283 [==============================] - 6s 19ms/step - loss: 1.3555 - accuracy: 0.5923 - val_loss: 1.4208 - val_accuracy: 0.5620
Epoch 109/150
283/283 [==============================] - 6s 18ms/step - loss: 1.3616 - accuracy: 0.5914 - val_loss: 1.4099 - val_accuracy: 0.5803
Epoch 110/150
283/283 [==============================] - 6s 18ms/step - loss: 1.3580 - accuracy: 0.5938 - val_loss: 1.4482 - val_accuracy: 0.5520
Epoch 111/150
283/283 [==============================] - 5s 18ms/step - loss: 1.3493 - accuracy: 0.5948 - val_loss: 1.4680 - val_accuracy: 0.5600
Epoch 112/150
283/283 [==============================] - 6s 18ms/step - loss: 1.3473 - accuracy: 0.5965 - val_loss: 1.4038 - val_accuracy: 0.5787
Epoch 113/150
283/283 [==============================] - 6s 18ms/step - loss: 1.3412 - accuracy: 0.5946 - val_loss: 1.4693 - val_accuracy: 0.5633
Epoch 114/150
283/283 [==============================] - 6s 18ms/step - loss: 1.3398 - accuracy: 0.5916 - val_loss: 1.4027 - val_accuracy: 0.5967
Epoch 115/150
283/283 [==============================] - 5s 18ms/step - loss: 1.3357 - accuracy: 0.5961 - val_loss: 1.3977 - val_accuracy: 0.5907
Epoch 116/150
283/283 [==============================] - 5s 18ms/step - loss: 1.3354 - accuracy: 0.5980 - val_loss: 1.3975 - val_accuracy: 0.5740
Epoch 117/150
283/283 [==============================] - 5s 18ms/step - loss: 1.3235 - accuracy: 0.5997 - val_loss: 1.3747 - val_accuracy: 0.5790
Epoch 118/150
283/283 [==============================] - 6s 18ms/step - loss: 1.3284 - accuracy: 0.5957 - val_loss: 1.3873 - val_accuracy: 0.5787
Epoch 119/150
283/283 [==============================] - 6s 18ms/step - loss: 1.3274 - accuracy: 0.6014 - val_loss: 1.4009 - val_accuracy: 0.5833
Epoch 120/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3214 - accuracy: 0.6015 - val_loss: 1.3758 - val_accuracy: 0.6000
Epoch 121/150
283/283 [==============================] - 5s 18ms/step - loss: 1.3139 - accuracy: 0.6076 - val_loss: 1.4082 - val_accuracy: 0.5683
Epoch 122/150
283/283 [==============================] - 5s 18ms/step - loss: 1.3171 - accuracy: 0.6045 - val_loss: 1.3891 - val_accuracy: 0.5763
Epoch 123/150
283/283 [==============================] - 6s 18ms/step - loss: 1.3110 - accuracy: 0.6039 - val_loss: 1.3606 - val_accuracy: 0.5923
Epoch 124/150
283/283 [==============================] - 5s 17ms/step - loss: 1.3130 - accuracy: 0.6043 - val_loss: 1.3926 - val_accuracy: 0.5873
Epoch 125/150
283/283 [==============================] - 5s 18ms/step - loss: 1.3106 - accuracy: 0.6011 - val_loss: 1.3309 - val_accuracy: 0.6030
Epoch 126/150
283/283 [==============================] - 6s 19ms/step - loss: 1.3060 - accuracy: 0.6059 - val_loss: 1.3774 - val_accuracy: 0.5690
Epoch 127/150
283/283 [==============================] - 5s 18ms/step - loss: 1.3048 - accuracy: 0.6062 - val_loss: 1.3760 - val_accuracy: 0.5743
Epoch 128/150
283/283 [==============================] - 5s 18ms/step - loss: 1.3085 - accuracy: 0.6027 - val_loss: 1.3417 - val_accuracy: 0.6153
Epoch 129/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2916 - accuracy: 0.6099 - val_loss: 1.3631 - val_accuracy: 0.5973
Epoch 130/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2974 - accuracy: 0.6072 - val_loss: 1.3460 - val_accuracy: 0.5993
Epoch 131/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2929 - accuracy: 0.6105 - val_loss: 1.3467 - val_accuracy: 0.5903
Epoch 132/150
283/283 [==============================] - 6s 18ms/step - loss: 1.2818 - accuracy: 0.6136 - val_loss: 1.3232 - val_accuracy: 0.6053
Epoch 133/150
283/283 [==============================] - 6s 18ms/step - loss: 1.2931 - accuracy: 0.6100 - val_loss: 1.3426 - val_accuracy: 0.6007
Epoch 134/150
283/283 [==============================] - 6s 18ms/step - loss: 1.2829 - accuracy: 0.6092 - val_loss: 1.3649 - val_accuracy: 0.5767
Epoch 135/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2822 - accuracy: 0.6123 - val_loss: 1.3208 - val_accuracy: 0.5997
Epoch 136/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2799 - accuracy: 0.6101 - val_loss: 1.3000 - val_accuracy: 0.6197
Epoch 137/150
283/283 [==============================] - 6s 19ms/step - loss: 1.2728 - accuracy: 0.6166 - val_loss: 1.3418 - val_accuracy: 0.5973
Epoch 138/150
283/283 [==============================] - 6s 19ms/step - loss: 1.2818 - accuracy: 0.6187 - val_loss: 1.3629 - val_accuracy: 0.5737
Epoch 139/150
283/283 [==============================] - 6s 18ms/step - loss: 1.2746 - accuracy: 0.6141 - val_loss: 1.3287 - val_accuracy: 0.5877
Epoch 140/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2800 - accuracy: 0.6102 - val_loss: 1.3584 - val_accuracy: 0.5870
Epoch 141/150
283/283 [==============================] - 6s 18ms/step - loss: 1.2734 - accuracy: 0.6132 - val_loss: 1.3030 - val_accuracy: 0.6083
Epoch 142/150
283/283 [==============================] - 6s 19ms/step - loss: 1.2634 - accuracy: 0.6184 - val_loss: 1.3377 - val_accuracy: 0.6133
Epoch 143/150
283/283 [==============================] - 6s 18ms/step - loss: 1.2655 - accuracy: 0.6123 - val_loss: 1.3184 - val_accuracy: 0.5980
Epoch 144/150
283/283 [==============================] - 6s 19ms/step - loss: 1.2610 - accuracy: 0.6197 - val_loss: 1.3210 - val_accuracy: 0.6177
Epoch 145/150
283/283 [==============================] - 6s 19ms/step - loss: 1.2585 - accuracy: 0.6162 - val_loss: 1.2888 - val_accuracy: 0.6193
Epoch 146/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2584 - accuracy: 0.6224 - val_loss: 1.3954 - val_accuracy: 0.5757
Epoch 147/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2620 - accuracy: 0.6186 - val_loss: 1.2872 - val_accuracy: 0.5977
Epoch 148/150
283/283 [==============================] - 6s 19ms/step - loss: 1.2517 - accuracy: 0.6203 - val_loss: 1.3053 - val_accuracy: 0.5923
Epoch 149/150
283/283 [==============================] - 6s 19ms/step - loss: 1.2509 - accuracy: 0.6197 - val_loss: 1.2885 - val_accuracy: 0.6140
Epoch 150/150
283/283 [==============================] - 5s 17ms/step - loss: 1.2486 - accuracy: 0.6249 - val_loss: 1.3337 - val_accuracy: 0.6287

history = baseline128desaturation_model.fit(res128desaturation_train_data, validation_data = res128desaturation_validation_data, epochs = 150, callbacks = earlyStopper)
cnnf.class_image(res128desaturation_full_data, class_names, cmap = "gray")
cnnf.model_metric_graph(history, baseline128desaturation_model, res128desaturation_validation_data, name = "baseline model for 128x128 image (using desaturation method)", class_names = class_names)

Epoch 1/150
283/283 [==============================] - 7s 20ms/step - loss: 2.6102 - accuracy: 0.1426 - val_loss: 2.7569 - val_accuracy: 0.1147
Epoch 2/150
283/283 [==============================] - 6s 20ms/step - loss: 2.4965 - accuracy: 0.1989 - val_loss: 2.6472 - val_accuracy: 0.1853
Epoch 3/150
283/283 [==============================] - 6s 20ms/step - loss: 2.4159 - accuracy: 0.2409 - val_loss: 2.5997 - val_accuracy: 0.1730
Epoch 4/150
283/283 [==============================] - 6s 20ms/step - loss: 2.3631 - accuracy: 0.2584 - val_loss: 2.5061 - val_accuracy: 0.2063
Epoch 5/150
283/283 [==============================] - 6s 20ms/step - loss: 2.3123 - accuracy: 0.2720 - val_loss: 2.4435 - val_accuracy: 0.2197
Epoch 6/150
283/283 [==============================] - 6s 19ms/step - loss: 2.2634 - accuracy: 0.2812 - val_loss: 2.3851 - val_accuracy: 0.2160
Epoch 7/150
283/283 [==============================] - 6s 18ms/step - loss: 2.2193 - accuracy: 0.3036 - val_loss: 2.3640 - val_accuracy: 0.2280
Epoch 8/150
283/283 [==============================] - 6s 18ms/step - loss: 2.1774 - accuracy: 0.3286 - val_loss: 2.2765 - val_accuracy: 0.2630
Epoch 9/150
283/283 [==============================] - 6s 19ms/step - loss: 2.1427 - accuracy: 0.3385 - val_loss: 2.2628 - val_accuracy: 0.2840
Epoch 10/150
283/283 [==============================] - 5s 18ms/step - loss: 2.1080 - accuracy: 0.3470 - val_loss: 2.2673 - val_accuracy: 0.2740
Epoch 11/150
283/283 [==============================] - 5s 18ms/step - loss: 2.0877 - accuracy: 0.3573 - val_loss: 2.2353 - val_accuracy: 0.2953
Epoch 12/150
283/283 [==============================] - 6s 19ms/step - loss: 2.0659 - accuracy: 0.3556 - val_loss: 2.1905 - val_accuracy: 0.2950
Epoch 13/150
283/283 [==============================] - 5s 18ms/step - loss: 2.0458 - accuracy: 0.3661 - val_loss: 2.1435 - val_accuracy: 0.3097
Epoch 14/150
283/283 [==============================] - 6s 20ms/step - loss: 2.0367 - accuracy: 0.3694 - val_loss: 2.1633 - val_accuracy: 0.3063
Epoch 15/150
283/283 [==============================] - 6s 19ms/step - loss: 2.0195 - accuracy: 0.3766 - val_loss: 2.1196 - val_accuracy: 0.3070
Epoch 16/150
283/283 [==============================] - 6s 19ms/step - loss: 2.0152 - accuracy: 0.3747 - val_loss: 2.0998 - val_accuracy: 0.3157
Epoch 17/150
283/283 [==============================] - 6s 19ms/step - loss: 1.9993 - accuracy: 0.3732 - val_loss: 2.0893 - val_accuracy: 0.3337
Epoch 18/150
283/283 [==============================] - 6s 19ms/step - loss: 1.9909 - accuracy: 0.3820 - val_loss: 2.1594 - val_accuracy: 0.3087
Epoch 19/150
283/283 [==============================] - 6s 19ms/step - loss: 1.9802 - accuracy: 0.3842 - val_loss: 2.0845 - val_accuracy: 0.3257
Epoch 20/150
283/283 [==============================] - 6s 18ms/step - loss: 1.9749 - accuracy: 0.3881 - val_loss: 2.0641 - val_accuracy: 0.3287
Epoch 21/150
283/283 [==============================] - 5s 18ms/step - loss: 1.9631 - accuracy: 0.3922 - val_loss: 2.0569 - val_accuracy: 0.3257
Epoch 22/150
283/283 [==============================] - 6s 18ms/step - loss: 1.9545 - accuracy: 0.3889 - val_loss: 2.0403 - val_accuracy: 0.3310
Epoch 23/150
283/283 [==============================] - 6s 18ms/step - loss: 1.9425 - accuracy: 0.3949 - val_loss: 2.0359 - val_accuracy: 0.3403
Epoch 24/150
283/283 [==============================] - 5s 18ms/step - loss: 1.9336 - accuracy: 0.3992 - val_loss: 2.0372 - val_accuracy: 0.3447
Epoch 25/150
283/283 [==============================] - 6s 18ms/step - loss: 1.9222 - accuracy: 0.4082 - val_loss: 2.0732 - val_accuracy: 0.3400
Epoch 26/150
283/283 [==============================] - 6s 19ms/step - loss: 1.9185 - accuracy: 0.3996 - val_loss: 2.1175 - val_accuracy: 0.3237
Epoch 27/150
283/283 [==============================] - 6s 19ms/step - loss: 1.9039 - accuracy: 0.4102 - val_loss: 2.0213 - val_accuracy: 0.3597
Epoch 28/150
283/283 [==============================] - 6s 19ms/step - loss: 1.8989 - accuracy: 0.4092 - val_loss: 2.0140 - val_accuracy: 0.3530
Epoch 29/150
283/283 [==============================] - 6s 18ms/step - loss: 1.8837 - accuracy: 0.4173 - val_loss: 1.9982 - val_accuracy: 0.3643
Epoch 30/150
283/283 [==============================] - 6s 19ms/step - loss: 1.8793 - accuracy: 0.4226 - val_loss: 2.0238 - val_accuracy: 0.3577
Epoch 31/150
283/283 [==============================] - 6s 18ms/step - loss: 1.8620 - accuracy: 0.4274 - val_loss: 1.9689 - val_accuracy: 0.3723
Epoch 32/150
283/283 [==============================] - 6s 19ms/step - loss: 1.8566 - accuracy: 0.4278 - val_loss: 1.9579 - val_accuracy: 0.3720
Epoch 33/150
283/283 [==============================] - 6s 19ms/step - loss: 1.8434 - accuracy: 0.4304 - val_loss: 1.9803 - val_accuracy: 0.3637
Epoch 34/150
283/283 [==============================] - 5s 18ms/step - loss: 1.8364 - accuracy: 0.4339 - val_loss: 1.9402 - val_accuracy: 0.3733
Epoch 35/150
283/283 [==============================] - 6s 18ms/step - loss: 1.8198 - accuracy: 0.4423 - val_loss: 1.9816 - val_accuracy: 0.3707
Epoch 36/150
283/283 [==============================] - 6s 19ms/step - loss: 1.8067 - accuracy: 0.4426 - val_loss: 1.9196 - val_accuracy: 0.3743
Epoch 37/150
283/283 [==============================] - 6s 19ms/step - loss: 1.7969 - accuracy: 0.4448 - val_loss: 2.0591 - val_accuracy: 0.3483
Epoch 38/150
283/283 [==============================] - 6s 20ms/step - loss: 1.7882 - accuracy: 0.4505 - val_loss: 1.8716 - val_accuracy: 0.4007
Epoch 39/150
283/283 [==============================] - 6s 19ms/step - loss: 1.7722 - accuracy: 0.4559 - val_loss: 1.8928 - val_accuracy: 0.3963
Epoch 40/150
283/283 [==============================] - 6s 20ms/step - loss: 1.7635 - accuracy: 0.4567 - val_loss: 1.9044 - val_accuracy: 0.3973
Epoch 41/150
283/283 [==============================] - 6s 20ms/step - loss: 1.7506 - accuracy: 0.4626 - val_loss: 1.8362 - val_accuracy: 0.4050
Epoch 42/150
283/283 [==============================] - 6s 19ms/step - loss: 1.7427 - accuracy: 0.4659 - val_loss: 1.8671 - val_accuracy: 0.4050
Epoch 43/150
283/283 [==============================] - 6s 18ms/step - loss: 1.7249 - accuracy: 0.4667 - val_loss: 1.8166 - val_accuracy: 0.4120
Epoch 44/150
283/283 [==============================] - 6s 20ms/step - loss: 1.7165 - accuracy: 0.4745 - val_loss: 1.8433 - val_accuracy: 0.4050
Epoch 45/150
283/283 [==============================] - 6s 19ms/step - loss: 1.7023 - accuracy: 0.4759 - val_loss: 1.8286 - val_accuracy: 0.4197
Epoch 46/150
283/283 [==============================] - 6s 18ms/step - loss: 1.6924 - accuracy: 0.4784 - val_loss: 1.8084 - val_accuracy: 0.4247
Epoch 47/150
283/283 [==============================] - 6s 20ms/step - loss: 1.6792 - accuracy: 0.4865 - val_loss: 1.8430 - val_accuracy: 0.4143
Epoch 48/150
283/283 [==============================] - 6s 19ms/step - loss: 1.6671 - accuracy: 0.4909 - val_loss: 1.7885 - val_accuracy: 0.4200
Epoch 49/150
283/283 [==============================] - 6s 19ms/step - loss: 1.6584 - accuracy: 0.4900 - val_loss: 1.7715 - val_accuracy: 0.4290
Epoch 50/150
283/283 [==============================] - 6s 18ms/step - loss: 1.6567 - accuracy: 0.4948 - val_loss: 1.7766 - val_accuracy: 0.4283
Epoch 51/150
283/283 [==============================] - 5s 18ms/step - loss: 1.6395 - accuracy: 0.4950 - val_loss: 1.8398 - val_accuracy: 0.4093
Epoch 52/150
283/283 [==============================] - 6s 19ms/step - loss: 1.6324 - accuracy: 0.5022 - val_loss: 1.7525 - val_accuracy: 0.4437
Epoch 53/150
283/283 [==============================] - 6s 19ms/step - loss: 1.6215 - accuracy: 0.5027 - val_loss: 1.7649 - val_accuracy: 0.4413
Epoch 54/150
283/283 [==============================] - 6s 19ms/step - loss: 1.6083 - accuracy: 0.5069 - val_loss: 1.6994 - val_accuracy: 0.4610
Epoch 55/150
283/283 [==============================] - 6s 19ms/step - loss: 1.6003 - accuracy: 0.5071 - val_loss: 1.7128 - val_accuracy: 0.4540
Epoch 56/150
283/283 [==============================] - 6s 18ms/step - loss: 1.6015 - accuracy: 0.5054 - val_loss: 1.7264 - val_accuracy: 0.4490
Epoch 57/150
283/283 [==============================] - 6s 19ms/step - loss: 1.5829 - accuracy: 0.5148 - val_loss: 1.6680 - val_accuracy: 0.4657
Epoch 58/150
283/283 [==============================] - 6s 19ms/step - loss: 1.5742 - accuracy: 0.5125 - val_loss: 1.6950 - val_accuracy: 0.4590
Epoch 59/150
283/283 [==============================] - 6s 19ms/step - loss: 1.5798 - accuracy: 0.5124 - val_loss: 1.7131 - val_accuracy: 0.4313
Epoch 60/150
283/283 [==============================] - 6s 20ms/step - loss: 1.5683 - accuracy: 0.5160 - val_loss: 1.6784 - val_accuracy: 0.4677
Epoch 61/150
283/283 [==============================] - 6s 19ms/step - loss: 1.5537 - accuracy: 0.5222 - val_loss: 1.6850 - val_accuracy: 0.4640
Epoch 62/150
283/283 [==============================] - 6s 18ms/step - loss: 1.5529 - accuracy: 0.5200 - val_loss: 1.6416 - val_accuracy: 0.4810
Epoch 63/150
283/283 [==============================] - 6s 18ms/step - loss: 1.5389 - accuracy: 0.5233 - val_loss: 1.6832 - val_accuracy: 0.4483
Epoch 64/150
283/283 [==============================] - 6s 19ms/step - loss: 1.5399 - accuracy: 0.5268 - val_loss: 1.6924 - val_accuracy: 0.4620
Epoch 65/150
283/283 [==============================] - 6s 20ms/step - loss: 1.5288 - accuracy: 0.5305 - val_loss: 1.6469 - val_accuracy: 0.4733
Epoch 66/150
283/283 [==============================] - 6s 19ms/step - loss: 1.5219 - accuracy: 0.5287 - val_loss: 1.6156 - val_accuracy: 0.4793
Epoch 67/150
283/283 [==============================] - 6s 19ms/step - loss: 1.5168 - accuracy: 0.5341 - val_loss: 1.5978 - val_accuracy: 0.4900
Epoch 68/150
283/283 [==============================] - 6s 19ms/step - loss: 1.5190 - accuracy: 0.5316 - val_loss: 1.6261 - val_accuracy: 0.4813
Epoch 69/150
283/283 [==============================] - 6s 19ms/step - loss: 1.5079 - accuracy: 0.5362 - val_loss: 1.6210 - val_accuracy: 0.4640
Epoch 70/150
283/283 [==============================] - 6s 19ms/step - loss: 1.5058 - accuracy: 0.5343 - val_loss: 1.6225 - val_accuracy: 0.4680
Epoch 71/150
283/283 [==============================] - 6s 20ms/step - loss: 1.5005 - accuracy: 0.5323 - val_loss: 1.6263 - val_accuracy: 0.4837
Epoch 72/150
283/283 [==============================] - 6s 20ms/step - loss: 1.4924 - accuracy: 0.5371 - val_loss: 1.6071 - val_accuracy: 0.4923
Epoch 73/150
283/283 [==============================] - 6s 20ms/step - loss: 1.4891 - accuracy: 0.5387 - val_loss: 1.5888 - val_accuracy: 0.4923
Epoch 74/150
283/283 [==============================] - 6s 20ms/step - loss: 1.4874 - accuracy: 0.5447 - val_loss: 1.5919 - val_accuracy: 0.4923
Epoch 75/150
283/283 [==============================] - 6s 20ms/step - loss: 1.4810 - accuracy: 0.5456 - val_loss: 1.5826 - val_accuracy: 0.5043
Epoch 76/150
283/283 [==============================] - 6s 20ms/step - loss: 1.4840 - accuracy: 0.5415 - val_loss: 1.6069 - val_accuracy: 0.4680
Epoch 77/150
283/283 [==============================] - 6s 20ms/step - loss: 1.4725 - accuracy: 0.5420 - val_loss: 1.5421 - val_accuracy: 0.5087
Epoch 78/150
283/283 [==============================] - 6s 21ms/step - loss: 1.4657 - accuracy: 0.5438 - val_loss: 1.5525 - val_accuracy: 0.5033
Epoch 79/150
283/283 [==============================] - 6s 20ms/step - loss: 1.4614 - accuracy: 0.5474 - val_loss: 1.5563 - val_accuracy: 0.4980
Epoch 80/150
283/283 [==============================] - 6s 20ms/step - loss: 1.4541 - accuracy: 0.5477 - val_loss: 1.5777 - val_accuracy: 0.4987
Epoch 81/150
283/283 [==============================] - 6s 18ms/step - loss: 1.4488 - accuracy: 0.5484 - val_loss: 1.5623 - val_accuracy: 0.4840
Epoch 82/150
283/283 [==============================] - 6s 19ms/step - loss: 1.4514 - accuracy: 0.5533 - val_loss: 1.5643 - val_accuracy: 0.5047
Epoch 83/150
283/283 [==============================] - 6s 19ms/step - loss: 1.4515 - accuracy: 0.5461 - val_loss: 1.5676 - val_accuracy: 0.4897
Epoch 84/150
283/283 [==============================] - 6s 19ms/step - loss: 1.4472 - accuracy: 0.5533 - val_loss: 1.6115 - val_accuracy: 0.4690
Epoch 85/150
283/283 [==============================] - 6s 18ms/step - loss: 1.4319 - accuracy: 0.5589 - val_loss: 1.5467 - val_accuracy: 0.4910
Epoch 86/150
283/283 [==============================] - 6s 19ms/step - loss: 1.4298 - accuracy: 0.5551 - val_loss: 1.4988 - val_accuracy: 0.5203
Epoch 87/150
283/283 [==============================] - 6s 19ms/step - loss: 1.4262 - accuracy: 0.5542 - val_loss: 1.5914 - val_accuracy: 0.4887
Epoch 88/150
283/283 [==============================] - 6s 19ms/step - loss: 1.4214 - accuracy: 0.5601 - val_loss: 1.5237 - val_accuracy: 0.5143
Epoch 89/150
283/283 [==============================] - 6s 19ms/step - loss: 1.4171 - accuracy: 0.5593 - val_loss: 1.5236 - val_accuracy: 0.5070
Epoch 90/150
283/283 [==============================] - 6s 20ms/step - loss: 1.4162 - accuracy: 0.5595 - val_loss: 1.5430 - val_accuracy: 0.5050
Epoch 91/150
283/283 [==============================] - 6s 20ms/step - loss: 1.4066 - accuracy: 0.5673 - val_loss: 1.4819 - val_accuracy: 0.5257
Epoch 92/150
283/283 [==============================] - 6s 19ms/step - loss: 1.4068 - accuracy: 0.5619 - val_loss: 1.5051 - val_accuracy: 0.5073
Epoch 93/150
283/283 [==============================] - 6s 19ms/step - loss: 1.4112 - accuracy: 0.5596 - val_loss: 1.5007 - val_accuracy: 0.5167
Epoch 94/150
283/283 [==============================] - 6s 19ms/step - loss: 1.3997 - accuracy: 0.5629 - val_loss: 1.4935 - val_accuracy: 0.5310
Epoch 95/150
283/283 [==============================] - 6s 20ms/step - loss: 1.4011 - accuracy: 0.5656 - val_loss: 1.4748 - val_accuracy: 0.5363
Epoch 96/150
283/283 [==============================] - 6s 20ms/step - loss: 1.3906 - accuracy: 0.5673 - val_loss: 1.4938 - val_accuracy: 0.5107
Epoch 97/150
283/283 [==============================] - 6s 21ms/step - loss: 1.3998 - accuracy: 0.5614 - val_loss: 1.4726 - val_accuracy: 0.5307
Epoch 98/150
283/283 [==============================] - 7s 22ms/step - loss: 1.3908 - accuracy: 0.5662 - val_loss: 1.4529 - val_accuracy: 0.5413
Epoch 99/150
283/283 [==============================] - 7s 22ms/step - loss: 1.3828 - accuracy: 0.5738 - val_loss: 1.4762 - val_accuracy: 0.5230
Epoch 100/150
283/283 [==============================] - 7s 22ms/step - loss: 1.3779 - accuracy: 0.5708 - val_loss: 1.4524 - val_accuracy: 0.5377
Epoch 101/150
283/283 [==============================] - 6s 20ms/step - loss: 1.3733 - accuracy: 0.5755 - val_loss: 1.4700 - val_accuracy: 0.5320
Epoch 102/150
283/283 [==============================] - 6s 19ms/step - loss: 1.3751 - accuracy: 0.5718 - val_loss: 1.4555 - val_accuracy: 0.5253
Epoch 103/150
283/283 [==============================] - 6s 20ms/step - loss: 1.3670 - accuracy: 0.5755 - val_loss: 1.4904 - val_accuracy: 0.5060
Epoch 104/150
283/283 [==============================] - 6s 19ms/step - loss: 1.3630 - accuracy: 0.5726 - val_loss: 1.4542 - val_accuracy: 0.5410
Epoch 105/150
283/283 [==============================] - 6s 20ms/step - loss: 1.3666 - accuracy: 0.5753 - val_loss: 1.4691 - val_accuracy: 0.5313
Epoch 106/150
283/283 [==============================] - 8s 26ms/step - loss: 1.3562 - accuracy: 0.5771 - val_loss: 1.4281 - val_accuracy: 0.5393
Epoch 107/150
283/283 [==============================] - 7s 24ms/step - loss: 1.3525 - accuracy: 0.5812 - val_loss: 1.4629 - val_accuracy: 0.5293
Epoch 108/150
283/283 [==============================] - 6s 21ms/step - loss: 1.3495 - accuracy: 0.5834 - val_loss: 1.4649 - val_accuracy: 0.5287
Epoch 109/150
283/283 [==============================] - 6s 21ms/step - loss: 1.3426 - accuracy: 0.5784 - val_loss: 1.5098 - val_accuracy: 0.4933
Epoch 110/150
283/283 [==============================] - 6s 20ms/step - loss: 1.3541 - accuracy: 0.5785 - val_loss: 1.4426 - val_accuracy: 0.5323
Epoch 111/150
283/283 [==============================] - 6s 21ms/step - loss: 1.3405 - accuracy: 0.5813 - val_loss: 1.4394 - val_accuracy: 0.5350
Epoch 112/150
283/283 [==============================] - 6s 21ms/step - loss: 1.3362 - accuracy: 0.5822 - val_loss: 1.4045 - val_accuracy: 0.5577
Epoch 113/150
283/283 [==============================] - 6s 19ms/step - loss: 1.3393 - accuracy: 0.5801 - val_loss: 1.4593 - val_accuracy: 0.5550
Epoch 114/150
283/283 [==============================] - 6s 20ms/step - loss: 1.3353 - accuracy: 0.5864 - val_loss: 1.4318 - val_accuracy: 0.5580
Epoch 115/150
283/283 [==============================] - 6s 20ms/step - loss: 1.3331 - accuracy: 0.5850 - val_loss: 1.5539 - val_accuracy: 0.4793
Epoch 116/150
283/283 [==============================] - 6s 20ms/step - loss: 1.3326 - accuracy: 0.5824 - val_loss: 1.4167 - val_accuracy: 0.5550
Epoch 117/150
283/283 [==============================] - 6s 20ms/step - loss: 1.3224 - accuracy: 0.5866 - val_loss: 1.4175 - val_accuracy: 0.5493
Epoch 118/150
283/283 [==============================] - 6s 21ms/step - loss: 1.3198 - accuracy: 0.5905 - val_loss: 1.3844 - val_accuracy: 0.5633
Epoch 119/150
283/283 [==============================] - 6s 20ms/step - loss: 1.3233 - accuracy: 0.5904 - val_loss: 1.3924 - val_accuracy: 0.5630
Epoch 120/150
283/283 [==============================] - 6s 21ms/step - loss: 1.3162 - accuracy: 0.5896 - val_loss: 1.4084 - val_accuracy: 0.5383
Epoch 121/150
283/283 [==============================] - 6s 20ms/step - loss: 1.3054 - accuracy: 0.5975 - val_loss: 1.3621 - val_accuracy: 0.5723
Epoch 122/150
283/283 [==============================] - 6s 20ms/step - loss: 1.3085 - accuracy: 0.5901 - val_loss: 1.4888 - val_accuracy: 0.5053
Epoch 123/150
283/283 [==============================] - 6s 20ms/step - loss: 1.3183 - accuracy: 0.5875 - val_loss: 1.3833 - val_accuracy: 0.5730
Epoch 124/150
283/283 [==============================] - 6s 20ms/step - loss: 1.3001 - accuracy: 0.5976 - val_loss: 1.4048 - val_accuracy: 0.5563
Epoch 125/150
283/283 [==============================] - 6s 20ms/step - loss: 1.2946 - accuracy: 0.5958 - val_loss: 1.3901 - val_accuracy: 0.5660
Epoch 126/150
283/283 [==============================] - 6s 21ms/step - loss: 1.2986 - accuracy: 0.5979 - val_loss: 1.3900 - val_accuracy: 0.5463
Epoch 127/150
283/283 [==============================] - 6s 20ms/step - loss: 1.2943 - accuracy: 0.5992 - val_loss: 1.4133 - val_accuracy: 0.5467
Epoch 128/150
283/283 [==============================] - 6s 20ms/step - loss: 1.2919 - accuracy: 0.5994 - val_loss: 1.3576 - val_accuracy: 0.5693
Epoch 129/150
283/283 [==============================] - 6s 20ms/step - loss: 1.2932 - accuracy: 0.5973 - val_loss: 1.3512 - val_accuracy: 0.5750
Epoch 130/150
283/283 [==============================] - 6s 21ms/step - loss: 1.2860 - accuracy: 0.5959 - val_loss: 1.3800 - val_accuracy: 0.5613
Epoch 131/150
283/283 [==============================] - 6s 19ms/step - loss: 1.2833 - accuracy: 0.6042 - val_loss: 1.5212 - val_accuracy: 0.5180
Epoch 132/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2843 - accuracy: 0.5958 - val_loss: 1.3550 - val_accuracy: 0.5783
Epoch 133/150
283/283 [==============================] - 6s 20ms/step - loss: 1.2728 - accuracy: 0.6031 - val_loss: 1.3965 - val_accuracy: 0.5647
Epoch 134/150
283/283 [==============================] - 5s 18ms/step - loss: 1.2671 - accuracy: 0.6061 - val_loss: 1.3845 - val_accuracy: 0.5560
Epoch 135/150
283/283 [==============================] - 6s 19ms/step - loss: 1.2789 - accuracy: 0.5984 - val_loss: 1.3724 - val_accuracy: 0.5573
Epoch 136/150
283/283 [==============================] - 6s 18ms/step - loss: 1.2713 - accuracy: 0.5989 - val_loss: 1.3627 - val_accuracy: 0.5557
Epoch 137/150
283/283 [==============================] - 6s 20ms/step - loss: 1.2643 - accuracy: 0.6030 - val_loss: 1.3537 - val_accuracy: 0.5737
Epoch 138/150
283/283 [==============================] - 6s 19ms/step - loss: 1.2703 - accuracy: 0.6043 - val_loss: 1.3661 - val_accuracy: 0.5757
Epoch 139/150
283/283 [==============================] - 6s 19ms/step - loss: 1.2646 - accuracy: 0.6015 - val_loss: 1.3388 - val_accuracy: 0.5780
Epoch 140/150
283/283 [==============================] - 6s 19ms/step - loss: 1.2547 - accuracy: 0.6072 - val_loss: 1.3191 - val_accuracy: 0.5790
Epoch 141/150
283/283 [==============================] - 6s 20ms/step - loss: 1.2556 - accuracy: 0.6061 - val_loss: 1.3292 - val_accuracy: 0.5733
Epoch 142/150
283/283 [==============================] - 6s 19ms/step - loss: 1.2515 - accuracy: 0.6091 - val_loss: 1.3956 - val_accuracy: 0.5443
Epoch 143/150
283/283 [==============================] - 6s 19ms/step - loss: 1.2570 - accuracy: 0.6045 - val_loss: 1.3551 - val_accuracy: 0.5613
Epoch 144/150
283/283 [==============================] - 6s 20ms/step - loss: 1.2469 - accuracy: 0.6057 - val_loss: 1.3258 - val_accuracy: 0.5820
Epoch 145/150
283/283 [==============================] - 6s 20ms/step - loss: 1.2403 - accuracy: 0.6167 - val_loss: 1.3406 - val_accuracy: 0.5597
Epoch 146/150
283/283 [==============================] - 6s 20ms/step - loss: 1.2382 - accuracy: 0.6159 - val_loss: 1.3075 - val_accuracy: 0.5807
Epoch 147/150
283/283 [==============================] - 6s 20ms/step - loss: 1.2416 - accuracy: 0.6107 - val_loss: 1.3137 - val_accuracy: 0.5940
Epoch 148/150
283/283 [==============================] - 6s 20ms/step - loss: 1.2443 - accuracy: 0.6083 - val_loss: 1.3479 - val_accuracy: 0.5753
Epoch 149/150
283/283 [==============================] - 6s 20ms/step - loss: 1.2391 - accuracy: 0.6153 - val_loss: 1.3397 - val_accuracy: 0.5690
Epoch 150/150
283/283 [==============================] - 6s 20ms/step - loss: 1.2366 - accuracy: 0.6134 - val_loss: 1.3032 - val_accuracy: 0.5860

---

Greyscale Conclusion (31 x 31)

---

res31_model_list = [baseline31_model, baseline31luminosity_model, baseline31desaturation_model]
res31_data_list = [res31greyscale_validation_data, res31luminosity_validation_data, res31desaturation_validation_data]
name_list = ["Normally Greyscaled", "Luminosity Method", "Desaturation Method"]
cnnf.model_comparison(res31_model_list, res31_data_list, name_list)

94/94 [==============================] - 1s 10ms/step - loss: 1.3386 - accuracy: 0.5730
94/94 [==============================] - 1s 10ms/step - loss: 1.2177 - accuracy: 0.6267
94/94 [==============================] - 1s 11ms/step - loss: 1.1963 - accuracy: 0.6060

From the above, we can see that for the 31 x 31 image size, the Luminosity Method is the best to use, yielding the highest overall classification accuracy, and the second-best model loss. Hence, for the 31 x 31 image size, we shall use the images greyscaled with the Luminosity method.

---

Greyscale Conclusion (128 x 128)

---

res128_model_list = [baseline128_model, baseline128luminosity_model, baseline128desaturation_model]
res128_data_list = [res128greyscale_validation_data, res128luminosity_validation_data, res128desaturation_validation_data]
name_list = ["Normally Greyscaled", "Luminosity Method", "Desaturation Method"]
cnnf.model_comparison(res128_model_list, res128_data_list, name_list)

94/94 [==============================] - 5s 37ms/step - loss: 1.3631 - accuracy: 0.5850
94/94 [==============================] - 1s 11ms/step - loss: 1.3337 - accuracy: 0.6287
94/94 [==============================] - 2s 14ms/step - loss: 1.3032 - accuracy: 0.5860

From the above comparison, we can see that for the 128 x 128 image size, the Luminosity Method is the best to use, as it gives the highest overall classification accuracy, as well as the second-best overall loss. Hence, we shall continue with the 128 x 128 image size using data greyscaled by the Luminosity Method.

Visualization of baseline model for 31 x 31 images.

tf.keras.utils.plot_model(baseline31_model, show_shapes=True, dpi=80)

tf.keras.utils.plot_model(baseline128_model, show_shapes=True, dpi=80)

---

Finalizing Data

---

Now, we shall normalize the final greyscaled data. Earlier on in the dataset, we did some basic normalization, namely, dividing it by 255. However, we shall now go a step further and shift the values, obtaining a standard deviation of 1, and a mean of 0.5 using Standardization. The formula for standardization is as follows: $$ \text Normalized\enspace Pixel\enspace Value = \frac{Pixel Value - \mu}{\theta} $$

# Function to compute mean and standard deviation for a dataset
def compute_mean_std(dataset, batch_size=32):
    # Initialize sum and sum of squares
    sum_images = 0
    sum_sq_images = 0
    num_samples = 0

    for images, _ in dataset.batch(batch_size):
        sum_images += tf.reduce_sum(images, axis=[0, 1, 2])
        sum_sq_images += tf.reduce_sum(tf.square(images), axis=[0, 1, 2])
        num_samples += tf.cast(tf.reduce_prod(tf.shape(images)[0:3]), tf.float32)

    # Compute mean and variance
    mean = sum_images / num_samples
    variance = (sum_sq_images / num_samples) - tf.square(mean)
    stddev = tf.sqrt(variance + 1e-7)

    return mean, stddev
# Function to standardize images
def standardize_image(image, label, mean, stddev):
    return (image - mean) / (stddev + 1e-7), label

# Compute mean and stddev for each dataset
# Make sure your datasets are not batched here
mean_res31, stddev_res31 = compute_mean_std(res31luminosity_train_data.unbatch())
mean_res128, stddev_res128 = compute_mean_std(res128luminosity_train_data.unbatch())

# Apply standardization to the datasets
std_res31_train = res31luminosity_train_data.map(
    lambda image, label: standardize_image(image, label, mean_res31, stddev_res31), 
    num_parallel_calls=tf.data.AUTOTUNE
)
std_res31_validation = res31luminosity_validation_data.map(
    lambda image, label: standardize_image(image, label, mean_res31, stddev_res31), 
    num_parallel_calls=tf.data.AUTOTUNE
)
std_res31_test = res31luminosity_test_data.map(
    lambda image, label: standardize_image(image, label, mean_res31, stddev_res31), 
    num_parallel_calls=tf.data.AUTOTUNE
)

std_res31_full = std_res31_train.concatenate(std_res31_validation)
std_res31_full = std_res31_full.concatenate(std_res31_test)

# Apply standardization to the res128 datasets
std_res128_train = res128luminosity_train_data.map(
    lambda image, label: standardize_image(image, label, mean_res128, stddev_res128), 
    num_parallel_calls=tf.data.AUTOTUNE
)
std_res128_validation = res128luminosity_validation_data.map(
    lambda image, label: standardize_image(image, label, mean_res128, stddev_res128), 
    num_parallel_calls=tf.data.AUTOTUNE
)
std_res128_test = res128luminosity_test_data.map(
    lambda image, label: standardize_image(image, label, mean_res128, stddev_res128), 
    num_parallel_calls=tf.data.AUTOTUNE
)

std_res128_full = std_res128_train.concatenate(std_res128_validation)
std_res128_full = std_res128_full.concatenate(std_res128_test)

for image, label in res31luminosity_train_data.unbatch().take(1):
    original_image = image.numpy()
    normalized_image, label = standardize_image(image, label, mean_res31, stddev_res31)
    normalized_image = normalized_image.numpy()

    # Plotting the original and normalized images
    plt.figure(figsize=(12, 6))

    plt.subplot(1, 2, 1)
    plt.imshow(original_image, cmap = "gray")
    plt.title("Original Image")
    plt.axis('off')

    plt.subplot(1, 2, 2)
    plt.imshow(normalized_image, cmap = "gray")
    plt.title("Normalized Image")
    plt.axis('off')

    plt.show()

for image, label in res128luminosity_train_data.unbatch().take(1):
    original_image = image.numpy()
    normalized_image, label = standardize_image(image, label, mean_res31, stddev_res31)
    normalized_image = normalized_image.numpy()

    # Plotting the original and normalized images
    plt.figure(figsize=(12, 6))

    plt.subplot(1, 2, 1)
    plt.imshow(original_image, cmap = "gray")
    plt.title("Original Image")
    plt.axis('off')

    plt.subplot(1, 2, 2)
    plt.imshow(normalized_image, cmap = "gray")
    plt.title("Normalized Image")
    plt.axis('off')

    plt.show()

cnnf.average_images_per_class(std_res31_full.unbatch(), class_names)

cnnf.average_images_per_class(std_res128_full.unbatch(), class_names)

From the above 4 plots, we can see that there is a slight difference after normalizing the images. The reason for the difference being this small may be due to the original mean already being close to the target of 0.5, hence requiring minimal shift in the pixel's brightness. We can also see that there are some differences in the features between the classes, with some having black centers, and some having whites. Some of the classes also have a "spot" in the middle, while some have the black/white spread through out the image. This tells us that there are potential features for the later models to pick apart, helping it to identify the individual classees.

---

Initial Modelling

---

Given that we already started using a basic CNN as a baseline model earlier, we shall now move onto using more complicated models for our task. We shall base our more complicated models off of the already established VGG and InceptionNet architectures, modifying them slightly to fit our use cases. Furthermore, as these models learn quickly compared to the baseline models we created earlier, we shall increase the patience of our EarlyStopper, to give the models more time to learn.

earlyStopper = EarlyStopping(monitor = 'val_loss', patience = 30)

What is VGG?

---

VGG, also known as VGGNet, is a classical convolutional neural network architecture, and was developed to increase the depth of CNNs to enhance it's performance. It stands for *Visual Geometry Group* , and is a standard deep CNN architecture, with multiple layers. Some iterations of VGG include *VGG-16*, and *VGG-19*, which consist of 16 and 19 convolutional layers respectively. It has formed the basis of other well-performing object detection models, and has done better than most other models on ImageNet. A VGG model is typically called "VGG-x", with x representing the number of layers it supports. For our use case, we shall make use of a simplified VGG-13 model, and a lighter version of the VGG-16 model. This is as the model typically accepts 224 x 224 images, but since we are using images of 31 x 31 and 128 x 128 respectively, we shall simplify the models to represent that.

A VGG model's architecture typically consists of an Input layer, followed by Convolutional Layers, Hidden Layers, then closing with Fully-Connected layers. A standard VGG-16 model would have 13 Convolutional Layers, and 3 Fully Connected Layers, meaning that it is a relatively extenstive network, with ~138 million trainable parameters. In addition, the number of filters that the model looks at doubles per "block", typically starting at 64, then 128, onto 256, and using 512 in the final layers. The large amount of filters typically allow it to identify a large amount of features, however, this comes with the downside of making it a large network, and thus taking a large amount of time to train.


vgg16Model = VGG16(
    include_top=True,
    weights='imagenet',
    input_tensor=None,
    input_shape=None,
    pooling=None,
    classes=1000,
    classifier_activation='softmax'
)

vgg16Model.summary()

Model: "vgg16"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 input_15 (InputLayer)       [(None, 224, 224, 3)]     0         
                                                                 
 block1_conv1 (Conv2D)       (None, 224, 224, 64)      1792      
                                                                 
 block1_conv2 (Conv2D)       (None, 224, 224, 64)      36928     
                                                                 
 block1_pool (MaxPooling2D)  (None, 112, 112, 64)      0         
                                                                 
 block2_conv1 (Conv2D)       (None, 112, 112, 128)     73856     
                                                                 
 block2_conv2 (Conv2D)       (None, 112, 112, 128)     147584    
                                                                 
 block2_pool (MaxPooling2D)  (None, 56, 56, 128)       0         
                                                                 
 block3_conv1 (Conv2D)       (None, 56, 56, 256)       295168    
                                                                 
 block3_conv2 (Conv2D)       (None, 56, 56, 256)       590080    
                                                                 
 block3_conv3 (Conv2D)       (None, 56, 56, 256)       590080    
                                                                 
 block3_pool (MaxPooling2D)  (None, 28, 28, 256)       0         
                                                                 
 block4_conv1 (Conv2D)       (None, 28, 28, 512)       1180160   
                                                                 
 block4_conv2 (Conv2D)       (None, 28, 28, 512)       2359808   
                                                                 
 block4_conv3 (Conv2D)       (None, 28, 28, 512)       2359808   
                                                                 
 block4_pool (MaxPooling2D)  (None, 14, 14, 512)       0         
                                                                 
 block5_conv1 (Conv2D)       (None, 14, 14, 512)       2359808   
                                                                 
 block5_conv2 (Conv2D)       (None, 14, 14, 512)       2359808   
                                                                 
 block5_conv3 (Conv2D)       (None, 14, 14, 512)       2359808   
                                                                 
 block5_pool (MaxPooling2D)  (None, 7, 7, 512)         0         
                                                                 
 flatten (Flatten)           (None, 25088)             0         
                                                                 
 fc1 (Dense)                 (None, 4096)              102764544 
                                                                 
 fc2 (Dense)                 (None, 4096)              16781312  
                                                                 
 predictions (Dense)         (None, 1000)              4097000   
                                                                 
=================================================================
Total params: 138,357,544
Trainable params: 138,357,544
Non-trainable params: 0
_________________________________________________________________

From the above, we can see that the example VGG16 model has 138 million parameters, with a 224 x 224 input layer. We shall simplify the model, and reduce the number of parameters to make it suitable for our 31 x 31 and 128 x 128 input images.

What is InceptionNet?

---

InceptionNet V3, also known as Inception V3, is an advanced convolutional neural network (CNN) architecture, notable for its innovative use of "Inception modules". It is a significant evolution of the original InceptionNet, or GoogLeNet, designed to increase the efficiency and accuracy of deep CNNs. This model is especially adept at reducing computational burden while enhancing the network's ability to learn rich and complex features from images. InceptionNet V3 features a complex arrangement of convolutional layers, pooling layers, and fully connected layers, alongside the characteristic Inception modules which involve parallel convolutions of different sizes.

The architecture of InceptionNet V3 includes an input layer, followed by a series of Inception modules, auxiliary classifiers for training, and ends with fully connected layers. This structure allows it to manage a more diverse range of filters simultaneously, enhancing its capability to recognize a wide array of features in images. While it maintains a high level of accuracy, Inception V3 is also optimized for efficient computation, making it a robust choice for challenging image recognition tasks. The model is known for its performance on ImageNet and other benchmark datasets, demonstrating its effectiveness in complex object detection and classification scenarios.

A typical InceptionNet V3 architecture consists of grid size reduction modules bewteen each version of "Inception Module". For our use case, we shall remove these grid-size reductions, given the already small size of our images, at 128 x 128, and 31 x 31, compared to the original 299 x 299 which InceptionNet V3 was created and optimized for.

inceptionModel = tf.keras.applications.InceptionV3(
    include_top=True,
    weights="imagenet",
    input_tensor=None,
    input_shape=None,
    pooling=None,
    classes=1000,
    classifier_activation="softmax",
)

inceptionModel.summary()

Model: "inception_v3"
__________________________________________________________________________________________________
 Layer (type)                   Output Shape         Param #     Connected to                     
==================================================================================================
 input_16 (InputLayer)          [(None, 299, 299, 3  0           []                               
                                )]                                                                
                                                                                                  
 conv2d_199 (Conv2D)            (None, 149, 149, 32  864         ['input_16[0][0]']               
                                )                                                                 
                                                                                                  
 batch_normalization_157 (Batch  (None, 149, 149, 32  96         ['conv2d_199[0][0]']             
 Normalization)                 )                                                                 
                                                                                                  
 activation (Activation)        (None, 149, 149, 32  0           ['batch_normalization_157[0][0]']
                                )                                                                 
                                                                                                  
 conv2d_200 (Conv2D)            (None, 147, 147, 32  9216        ['activation[0][0]']             
                                )                                                                 
                                                                                                  
 batch_normalization_158 (Batch  (None, 147, 147, 32  96         ['conv2d_200[0][0]']             
 Normalization)                 )                                                                 
                                                                                                  
 activation_1 (Activation)      (None, 147, 147, 32  0           ['batch_normalization_158[0][0]']
                                )                                                                 
                                                                                                  
 conv2d_201 (Conv2D)            (None, 147, 147, 64  18432       ['activation_1[0][0]']           
                                )                                                                 
                                                                                                  
 batch_normalization_159 (Batch  (None, 147, 147, 64  192        ['conv2d_201[0][0]']             
 Normalization)                 )                                                                 
                                                                                                  
 activation_2 (Activation)      (None, 147, 147, 64  0           ['batch_normalization_159[0][0]']
                                )                                                                 
                                                                                                  
 max_pooling2d_81 (MaxPooling2D  (None, 73, 73, 64)  0           ['activation_2[0][0]']           
 )                                                                                                
                                                                                                  
 conv2d_202 (Conv2D)            (None, 73, 73, 80)   5120        ['max_pooling2d_81[0][0]']       
                                                                                                  
 batch_normalization_160 (Batch  (None, 73, 73, 80)  240         ['conv2d_202[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_3 (Activation)      (None, 73, 73, 80)   0           ['batch_normalization_160[0][0]']
                                                                                                  
 conv2d_203 (Conv2D)            (None, 71, 71, 192)  138240      ['activation_3[0][0]']           
                                                                                                  
 batch_normalization_161 (Batch  (None, 71, 71, 192)  576        ['conv2d_203[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_4 (Activation)      (None, 71, 71, 192)  0           ['batch_normalization_161[0][0]']
                                                                                                  
 max_pooling2d_82 (MaxPooling2D  (None, 35, 35, 192)  0          ['activation_4[0][0]']           
 )                                                                                                
                                                                                                  
 conv2d_207 (Conv2D)            (None, 35, 35, 64)   12288       ['max_pooling2d_82[0][0]']       
                                                                                                  
 batch_normalization_165 (Batch  (None, 35, 35, 64)  192         ['conv2d_207[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_8 (Activation)      (None, 35, 35, 64)   0           ['batch_normalization_165[0][0]']
                                                                                                  
 conv2d_205 (Conv2D)            (None, 35, 35, 48)   9216        ['max_pooling2d_82[0][0]']       
                                                                                                  
 conv2d_208 (Conv2D)            (None, 35, 35, 96)   55296       ['activation_8[0][0]']           
                                                                                                  
 batch_normalization_163 (Batch  (None, 35, 35, 48)  144         ['conv2d_205[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_166 (Batch  (None, 35, 35, 96)  288         ['conv2d_208[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_6 (Activation)      (None, 35, 35, 48)   0           ['batch_normalization_163[0][0]']
                                                                                                  
 activation_9 (Activation)      (None, 35, 35, 96)   0           ['batch_normalization_166[0][0]']
                                                                                                  
 average_pooling2d_7 (AveragePo  (None, 35, 35, 192)  0          ['max_pooling2d_82[0][0]']       
 oling2D)                                                                                         
                                                                                                  
 conv2d_204 (Conv2D)            (None, 35, 35, 64)   12288       ['max_pooling2d_82[0][0]']       
                                                                                                  
 conv2d_206 (Conv2D)            (None, 35, 35, 64)   76800       ['activation_6[0][0]']           
                                                                                                  
 conv2d_209 (Conv2D)            (None, 35, 35, 96)   82944       ['activation_9[0][0]']           
                                                                                                  
 conv2d_210 (Conv2D)            (None, 35, 35, 32)   6144        ['average_pooling2d_7[0][0]']    
                                                                                                  
 batch_normalization_162 (Batch  (None, 35, 35, 64)  192         ['conv2d_204[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_164 (Batch  (None, 35, 35, 64)  192         ['conv2d_206[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_167 (Batch  (None, 35, 35, 96)  288         ['conv2d_209[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_168 (Batch  (None, 35, 35, 32)  96          ['conv2d_210[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_5 (Activation)      (None, 35, 35, 64)   0           ['batch_normalization_162[0][0]']
                                                                                                  
 activation_7 (Activation)      (None, 35, 35, 64)   0           ['batch_normalization_164[0][0]']
                                                                                                  
 activation_10 (Activation)     (None, 35, 35, 96)   0           ['batch_normalization_167[0][0]']
                                                                                                  
 activation_11 (Activation)     (None, 35, 35, 32)   0           ['batch_normalization_168[0][0]']
                                                                                                  
 mixed0 (Concatenate)           (None, 35, 35, 256)  0           ['activation_5[0][0]',           
                                                                  'activation_7[0][0]',           
                                                                  'activation_10[0][0]',          
                                                                  'activation_11[0][0]']          
                                                                                                  
 conv2d_214 (Conv2D)            (None, 35, 35, 64)   16384       ['mixed0[0][0]']                 
                                                                                                  
 batch_normalization_172 (Batch  (None, 35, 35, 64)  192         ['conv2d_214[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_15 (Activation)     (None, 35, 35, 64)   0           ['batch_normalization_172[0][0]']
                                                                                                  
 conv2d_212 (Conv2D)            (None, 35, 35, 48)   12288       ['mixed0[0][0]']                 
                                                                                                  
 conv2d_215 (Conv2D)            (None, 35, 35, 96)   55296       ['activation_15[0][0]']          
                                                                                                  
 batch_normalization_170 (Batch  (None, 35, 35, 48)  144         ['conv2d_212[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_173 (Batch  (None, 35, 35, 96)  288         ['conv2d_215[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_13 (Activation)     (None, 35, 35, 48)   0           ['batch_normalization_170[0][0]']
                                                                                                  
 activation_16 (Activation)     (None, 35, 35, 96)   0           ['batch_normalization_173[0][0]']
                                                                                                  
 average_pooling2d_8 (AveragePo  (None, 35, 35, 256)  0          ['mixed0[0][0]']                 
 oling2D)                                                                                         
                                                                                                  
 conv2d_211 (Conv2D)            (None, 35, 35, 64)   16384       ['mixed0[0][0]']                 
                                                                                                  
 conv2d_213 (Conv2D)            (None, 35, 35, 64)   76800       ['activation_13[0][0]']          
                                                                                                  
 conv2d_216 (Conv2D)            (None, 35, 35, 96)   82944       ['activation_16[0][0]']          
                                                                                                  
 conv2d_217 (Conv2D)            (None, 35, 35, 64)   16384       ['average_pooling2d_8[0][0]']    
                                                                                                  
 batch_normalization_169 (Batch  (None, 35, 35, 64)  192         ['conv2d_211[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_171 (Batch  (None, 35, 35, 64)  192         ['conv2d_213[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_174 (Batch  (None, 35, 35, 96)  288         ['conv2d_216[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_175 (Batch  (None, 35, 35, 64)  192         ['conv2d_217[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_12 (Activation)     (None, 35, 35, 64)   0           ['batch_normalization_169[0][0]']
                                                                                                  
 activation_14 (Activation)     (None, 35, 35, 64)   0           ['batch_normalization_171[0][0]']
                                                                                                  
 activation_17 (Activation)     (None, 35, 35, 96)   0           ['batch_normalization_174[0][0]']
                                                                                                  
 activation_18 (Activation)     (None, 35, 35, 64)   0           ['batch_normalization_175[0][0]']
                                                                                                  
 mixed1 (Concatenate)           (None, 35, 35, 288)  0           ['activation_12[0][0]',          
                                                                  'activation_14[0][0]',          
                                                                  'activation_17[0][0]',          
                                                                  'activation_18[0][0]']          
                                                                                                  
 conv2d_221 (Conv2D)            (None, 35, 35, 64)   18432       ['mixed1[0][0]']                 
                                                                                                  
 batch_normalization_179 (Batch  (None, 35, 35, 64)  192         ['conv2d_221[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_22 (Activation)     (None, 35, 35, 64)   0           ['batch_normalization_179[0][0]']
                                                                                                  
 conv2d_219 (Conv2D)            (None, 35, 35, 48)   13824       ['mixed1[0][0]']                 
                                                                                                  
 conv2d_222 (Conv2D)            (None, 35, 35, 96)   55296       ['activation_22[0][0]']          
                                                                                                  
 batch_normalization_177 (Batch  (None, 35, 35, 48)  144         ['conv2d_219[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_180 (Batch  (None, 35, 35, 96)  288         ['conv2d_222[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_20 (Activation)     (None, 35, 35, 48)   0           ['batch_normalization_177[0][0]']
                                                                                                  
 activation_23 (Activation)     (None, 35, 35, 96)   0           ['batch_normalization_180[0][0]']
                                                                                                  
 average_pooling2d_9 (AveragePo  (None, 35, 35, 288)  0          ['mixed1[0][0]']                 
 oling2D)                                                                                         
                                                                                                  
 conv2d_218 (Conv2D)            (None, 35, 35, 64)   18432       ['mixed1[0][0]']                 
                                                                                                  
 conv2d_220 (Conv2D)            (None, 35, 35, 64)   76800       ['activation_20[0][0]']          
                                                                                                  
 conv2d_223 (Conv2D)            (None, 35, 35, 96)   82944       ['activation_23[0][0]']          
                                                                                                  
 conv2d_224 (Conv2D)            (None, 35, 35, 64)   18432       ['average_pooling2d_9[0][0]']    
                                                                                                  
 batch_normalization_176 (Batch  (None, 35, 35, 64)  192         ['conv2d_218[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_178 (Batch  (None, 35, 35, 64)  192         ['conv2d_220[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_181 (Batch  (None, 35, 35, 96)  288         ['conv2d_223[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_182 (Batch  (None, 35, 35, 64)  192         ['conv2d_224[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_19 (Activation)     (None, 35, 35, 64)   0           ['batch_normalization_176[0][0]']
                                                                                                  
 activation_21 (Activation)     (None, 35, 35, 64)   0           ['batch_normalization_178[0][0]']
                                                                                                  
 activation_24 (Activation)     (None, 35, 35, 96)   0           ['batch_normalization_181[0][0]']
                                                                                                  
 activation_25 (Activation)     (None, 35, 35, 64)   0           ['batch_normalization_182[0][0]']
                                                                                                  
 mixed2 (Concatenate)           (None, 35, 35, 288)  0           ['activation_19[0][0]',          
                                                                  'activation_21[0][0]',          
                                                                  'activation_24[0][0]',          
                                                                  'activation_25[0][0]']          
                                                                                                  
 conv2d_226 (Conv2D)            (None, 35, 35, 64)   18432       ['mixed2[0][0]']                 
                                                                                                  
 batch_normalization_184 (Batch  (None, 35, 35, 64)  192         ['conv2d_226[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_27 (Activation)     (None, 35, 35, 64)   0           ['batch_normalization_184[0][0]']
                                                                                                  
 conv2d_227 (Conv2D)            (None, 35, 35, 96)   55296       ['activation_27[0][0]']          
                                                                                                  
 batch_normalization_185 (Batch  (None, 35, 35, 96)  288         ['conv2d_227[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_28 (Activation)     (None, 35, 35, 96)   0           ['batch_normalization_185[0][0]']
                                                                                                  
 conv2d_225 (Conv2D)            (None, 17, 17, 384)  995328      ['mixed2[0][0]']                 
                                                                                                  
 conv2d_228 (Conv2D)            (None, 17, 17, 96)   82944       ['activation_28[0][0]']          
                                                                                                  
 batch_normalization_183 (Batch  (None, 17, 17, 384)  1152       ['conv2d_225[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_186 (Batch  (None, 17, 17, 96)  288         ['conv2d_228[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_26 (Activation)     (None, 17, 17, 384)  0           ['batch_normalization_183[0][0]']
                                                                                                  
 activation_29 (Activation)     (None, 17, 17, 96)   0           ['batch_normalization_186[0][0]']
                                                                                                  
 max_pooling2d_83 (MaxPooling2D  (None, 17, 17, 288)  0          ['mixed2[0][0]']                 
 )                                                                                                
                                                                                                  
 mixed3 (Concatenate)           (None, 17, 17, 768)  0           ['activation_26[0][0]',          
                                                                  'activation_29[0][0]',          
                                                                  'max_pooling2d_83[0][0]']       
                                                                                                  
 conv2d_233 (Conv2D)            (None, 17, 17, 128)  98304       ['mixed3[0][0]']                 
                                                                                                  
 batch_normalization_191 (Batch  (None, 17, 17, 128)  384        ['conv2d_233[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_34 (Activation)     (None, 17, 17, 128)  0           ['batch_normalization_191[0][0]']
                                                                                                  
 conv2d_234 (Conv2D)            (None, 17, 17, 128)  114688      ['activation_34[0][0]']          
                                                                                                  
 batch_normalization_192 (Batch  (None, 17, 17, 128)  384        ['conv2d_234[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_35 (Activation)     (None, 17, 17, 128)  0           ['batch_normalization_192[0][0]']
                                                                                                  
 conv2d_230 (Conv2D)            (None, 17, 17, 128)  98304       ['mixed3[0][0]']                 
                                                                                                  
 conv2d_235 (Conv2D)            (None, 17, 17, 128)  114688      ['activation_35[0][0]']          
                                                                                                  
 batch_normalization_188 (Batch  (None, 17, 17, 128)  384        ['conv2d_230[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_193 (Batch  (None, 17, 17, 128)  384        ['conv2d_235[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_31 (Activation)     (None, 17, 17, 128)  0           ['batch_normalization_188[0][0]']
                                                                                                  
 activation_36 (Activation)     (None, 17, 17, 128)  0           ['batch_normalization_193[0][0]']
                                                                                                  
 conv2d_231 (Conv2D)            (None, 17, 17, 128)  114688      ['activation_31[0][0]']          
                                                                                                  
 conv2d_236 (Conv2D)            (None, 17, 17, 128)  114688      ['activation_36[0][0]']          
                                                                                                  
 batch_normalization_189 (Batch  (None, 17, 17, 128)  384        ['conv2d_231[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_194 (Batch  (None, 17, 17, 128)  384        ['conv2d_236[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_32 (Activation)     (None, 17, 17, 128)  0           ['batch_normalization_189[0][0]']
                                                                                                  
 activation_37 (Activation)     (None, 17, 17, 128)  0           ['batch_normalization_194[0][0]']
                                                                                                  
 average_pooling2d_10 (AverageP  (None, 17, 17, 768)  0          ['mixed3[0][0]']                 
 ooling2D)                                                                                        
                                                                                                  
 conv2d_229 (Conv2D)            (None, 17, 17, 192)  147456      ['mixed3[0][0]']                 
                                                                                                  
 conv2d_232 (Conv2D)            (None, 17, 17, 192)  172032      ['activation_32[0][0]']          
                                                                                                  
 conv2d_237 (Conv2D)            (None, 17, 17, 192)  172032      ['activation_37[0][0]']          
                                                                                                  
 conv2d_238 (Conv2D)            (None, 17, 17, 192)  147456      ['average_pooling2d_10[0][0]']   
                                                                                                  
 batch_normalization_187 (Batch  (None, 17, 17, 192)  576        ['conv2d_229[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_190 (Batch  (None, 17, 17, 192)  576        ['conv2d_232[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_195 (Batch  (None, 17, 17, 192)  576        ['conv2d_237[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_196 (Batch  (None, 17, 17, 192)  576        ['conv2d_238[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_30 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_187[0][0]']
                                                                                                  
 activation_33 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_190[0][0]']
                                                                                                  
 activation_38 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_195[0][0]']
                                                                                                  
 activation_39 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_196[0][0]']
                                                                                                  
 mixed4 (Concatenate)           (None, 17, 17, 768)  0           ['activation_30[0][0]',          
                                                                  'activation_33[0][0]',          
                                                                  'activation_38[0][0]',          
                                                                  'activation_39[0][0]']          
                                                                                                  
 conv2d_243 (Conv2D)            (None, 17, 17, 160)  122880      ['mixed4[0][0]']                 
                                                                                                  
 batch_normalization_201 (Batch  (None, 17, 17, 160)  480        ['conv2d_243[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_44 (Activation)     (None, 17, 17, 160)  0           ['batch_normalization_201[0][0]']
                                                                                                  
 conv2d_244 (Conv2D)            (None, 17, 17, 160)  179200      ['activation_44[0][0]']          
                                                                                                  
 batch_normalization_202 (Batch  (None, 17, 17, 160)  480        ['conv2d_244[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_45 (Activation)     (None, 17, 17, 160)  0           ['batch_normalization_202[0][0]']
                                                                                                  
 conv2d_240 (Conv2D)            (None, 17, 17, 160)  122880      ['mixed4[0][0]']                 
                                                                                                  
 conv2d_245 (Conv2D)            (None, 17, 17, 160)  179200      ['activation_45[0][0]']          
                                                                                                  
 batch_normalization_198 (Batch  (None, 17, 17, 160)  480        ['conv2d_240[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_203 (Batch  (None, 17, 17, 160)  480        ['conv2d_245[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_41 (Activation)     (None, 17, 17, 160)  0           ['batch_normalization_198[0][0]']
                                                                                                  
 activation_46 (Activation)     (None, 17, 17, 160)  0           ['batch_normalization_203[0][0]']
                                                                                                  
 conv2d_241 (Conv2D)            (None, 17, 17, 160)  179200      ['activation_41[0][0]']          
                                                                                                  
 conv2d_246 (Conv2D)            (None, 17, 17, 160)  179200      ['activation_46[0][0]']          
                                                                                                  
 batch_normalization_199 (Batch  (None, 17, 17, 160)  480        ['conv2d_241[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_204 (Batch  (None, 17, 17, 160)  480        ['conv2d_246[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_42 (Activation)     (None, 17, 17, 160)  0           ['batch_normalization_199[0][0]']
                                                                                                  
 activation_47 (Activation)     (None, 17, 17, 160)  0           ['batch_normalization_204[0][0]']
                                                                                                  
 average_pooling2d_11 (AverageP  (None, 17, 17, 768)  0          ['mixed4[0][0]']                 
 ooling2D)                                                                                        
                                                                                                  
 conv2d_239 (Conv2D)            (None, 17, 17, 192)  147456      ['mixed4[0][0]']                 
                                                                                                  
 conv2d_242 (Conv2D)            (None, 17, 17, 192)  215040      ['activation_42[0][0]']          
                                                                                                  
 conv2d_247 (Conv2D)            (None, 17, 17, 192)  215040      ['activation_47[0][0]']          
                                                                                                  
 conv2d_248 (Conv2D)            (None, 17, 17, 192)  147456      ['average_pooling2d_11[0][0]']   
                                                                                                  
 batch_normalization_197 (Batch  (None, 17, 17, 192)  576        ['conv2d_239[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_200 (Batch  (None, 17, 17, 192)  576        ['conv2d_242[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_205 (Batch  (None, 17, 17, 192)  576        ['conv2d_247[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_206 (Batch  (None, 17, 17, 192)  576        ['conv2d_248[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_40 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_197[0][0]']
                                                                                                  
 activation_43 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_200[0][0]']
                                                                                                  
 activation_48 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_205[0][0]']
                                                                                                  
 activation_49 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_206[0][0]']
                                                                                                  
 mixed5 (Concatenate)           (None, 17, 17, 768)  0           ['activation_40[0][0]',          
                                                                  'activation_43[0][0]',          
                                                                  'activation_48[0][0]',          
                                                                  'activation_49[0][0]']          
                                                                                                  
 conv2d_253 (Conv2D)            (None, 17, 17, 160)  122880      ['mixed5[0][0]']                 
                                                                                                  
 batch_normalization_211 (Batch  (None, 17, 17, 160)  480        ['conv2d_253[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_54 (Activation)     (None, 17, 17, 160)  0           ['batch_normalization_211[0][0]']
                                                                                                  
 conv2d_254 (Conv2D)            (None, 17, 17, 160)  179200      ['activation_54[0][0]']          
                                                                                                  
 batch_normalization_212 (Batch  (None, 17, 17, 160)  480        ['conv2d_254[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_55 (Activation)     (None, 17, 17, 160)  0           ['batch_normalization_212[0][0]']
                                                                                                  
 conv2d_250 (Conv2D)            (None, 17, 17, 160)  122880      ['mixed5[0][0]']                 
                                                                                                  
 conv2d_255 (Conv2D)            (None, 17, 17, 160)  179200      ['activation_55[0][0]']          
                                                                                                  
 batch_normalization_208 (Batch  (None, 17, 17, 160)  480        ['conv2d_250[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_213 (Batch  (None, 17, 17, 160)  480        ['conv2d_255[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_51 (Activation)     (None, 17, 17, 160)  0           ['batch_normalization_208[0][0]']
                                                                                                  
 activation_56 (Activation)     (None, 17, 17, 160)  0           ['batch_normalization_213[0][0]']
                                                                                                  
 conv2d_251 (Conv2D)            (None, 17, 17, 160)  179200      ['activation_51[0][0]']          
                                                                                                  
 conv2d_256 (Conv2D)            (None, 17, 17, 160)  179200      ['activation_56[0][0]']          
                                                                                                  
 batch_normalization_209 (Batch  (None, 17, 17, 160)  480        ['conv2d_251[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_214 (Batch  (None, 17, 17, 160)  480        ['conv2d_256[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_52 (Activation)     (None, 17, 17, 160)  0           ['batch_normalization_209[0][0]']
                                                                                                  
 activation_57 (Activation)     (None, 17, 17, 160)  0           ['batch_normalization_214[0][0]']
                                                                                                  
 average_pooling2d_12 (AverageP  (None, 17, 17, 768)  0          ['mixed5[0][0]']                 
 ooling2D)                                                                                        
                                                                                                  
 conv2d_249 (Conv2D)            (None, 17, 17, 192)  147456      ['mixed5[0][0]']                 
                                                                                                  
 conv2d_252 (Conv2D)            (None, 17, 17, 192)  215040      ['activation_52[0][0]']          
                                                                                                  
 conv2d_257 (Conv2D)            (None, 17, 17, 192)  215040      ['activation_57[0][0]']          
                                                                                                  
 conv2d_258 (Conv2D)            (None, 17, 17, 192)  147456      ['average_pooling2d_12[0][0]']   
                                                                                                  
 batch_normalization_207 (Batch  (None, 17, 17, 192)  576        ['conv2d_249[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_210 (Batch  (None, 17, 17, 192)  576        ['conv2d_252[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_215 (Batch  (None, 17, 17, 192)  576        ['conv2d_257[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_216 (Batch  (None, 17, 17, 192)  576        ['conv2d_258[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_50 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_207[0][0]']
                                                                                                  
 activation_53 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_210[0][0]']
                                                                                                  
 activation_58 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_215[0][0]']
                                                                                                  
 activation_59 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_216[0][0]']
                                                                                                  
 mixed6 (Concatenate)           (None, 17, 17, 768)  0           ['activation_50[0][0]',          
                                                                  'activation_53[0][0]',          
                                                                  'activation_58[0][0]',          
                                                                  'activation_59[0][0]']          
                                                                                                  
 conv2d_263 (Conv2D)            (None, 17, 17, 192)  147456      ['mixed6[0][0]']                 
                                                                                                  
 batch_normalization_221 (Batch  (None, 17, 17, 192)  576        ['conv2d_263[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_64 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_221[0][0]']
                                                                                                  
 conv2d_264 (Conv2D)            (None, 17, 17, 192)  258048      ['activation_64[0][0]']          
                                                                                                  
 batch_normalization_222 (Batch  (None, 17, 17, 192)  576        ['conv2d_264[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_65 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_222[0][0]']
                                                                                                  
 conv2d_260 (Conv2D)            (None, 17, 17, 192)  147456      ['mixed6[0][0]']                 
                                                                                                  
 conv2d_265 (Conv2D)            (None, 17, 17, 192)  258048      ['activation_65[0][0]']          
                                                                                                  
 batch_normalization_218 (Batch  (None, 17, 17, 192)  576        ['conv2d_260[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_223 (Batch  (None, 17, 17, 192)  576        ['conv2d_265[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_61 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_218[0][0]']
                                                                                                  
 activation_66 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_223[0][0]']
                                                                                                  
 conv2d_261 (Conv2D)            (None, 17, 17, 192)  258048      ['activation_61[0][0]']          
                                                                                                  
 conv2d_266 (Conv2D)            (None, 17, 17, 192)  258048      ['activation_66[0][0]']          
                                                                                                  
 batch_normalization_219 (Batch  (None, 17, 17, 192)  576        ['conv2d_261[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_224 (Batch  (None, 17, 17, 192)  576        ['conv2d_266[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_62 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_219[0][0]']
                                                                                                  
 activation_67 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_224[0][0]']
                                                                                                  
 average_pooling2d_13 (AverageP  (None, 17, 17, 768)  0          ['mixed6[0][0]']                 
 ooling2D)                                                                                        
                                                                                                  
 conv2d_259 (Conv2D)            (None, 17, 17, 192)  147456      ['mixed6[0][0]']                 
                                                                                                  
 conv2d_262 (Conv2D)            (None, 17, 17, 192)  258048      ['activation_62[0][0]']          
                                                                                                  
 conv2d_267 (Conv2D)            (None, 17, 17, 192)  258048      ['activation_67[0][0]']          
                                                                                                  
 conv2d_268 (Conv2D)            (None, 17, 17, 192)  147456      ['average_pooling2d_13[0][0]']   
                                                                                                  
 batch_normalization_217 (Batch  (None, 17, 17, 192)  576        ['conv2d_259[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_220 (Batch  (None, 17, 17, 192)  576        ['conv2d_262[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_225 (Batch  (None, 17, 17, 192)  576        ['conv2d_267[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_226 (Batch  (None, 17, 17, 192)  576        ['conv2d_268[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_60 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_217[0][0]']
                                                                                                  
 activation_63 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_220[0][0]']
                                                                                                  
 activation_68 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_225[0][0]']
                                                                                                  
 activation_69 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_226[0][0]']
                                                                                                  
 mixed7 (Concatenate)           (None, 17, 17, 768)  0           ['activation_60[0][0]',          
                                                                  'activation_63[0][0]',          
                                                                  'activation_68[0][0]',          
                                                                  'activation_69[0][0]']          
                                                                                                  
 conv2d_271 (Conv2D)            (None, 17, 17, 192)  147456      ['mixed7[0][0]']                 
                                                                                                  
 batch_normalization_229 (Batch  (None, 17, 17, 192)  576        ['conv2d_271[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_72 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_229[0][0]']
                                                                                                  
 conv2d_272 (Conv2D)            (None, 17, 17, 192)  258048      ['activation_72[0][0]']          
                                                                                                  
 batch_normalization_230 (Batch  (None, 17, 17, 192)  576        ['conv2d_272[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_73 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_230[0][0]']
                                                                                                  
 conv2d_269 (Conv2D)            (None, 17, 17, 192)  147456      ['mixed7[0][0]']                 
                                                                                                  
 conv2d_273 (Conv2D)            (None, 17, 17, 192)  258048      ['activation_73[0][0]']          
                                                                                                  
 batch_normalization_227 (Batch  (None, 17, 17, 192)  576        ['conv2d_269[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_231 (Batch  (None, 17, 17, 192)  576        ['conv2d_273[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_70 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_227[0][0]']
                                                                                                  
 activation_74 (Activation)     (None, 17, 17, 192)  0           ['batch_normalization_231[0][0]']
                                                                                                  
 conv2d_270 (Conv2D)            (None, 8, 8, 320)    552960      ['activation_70[0][0]']          
                                                                                                  
 conv2d_274 (Conv2D)            (None, 8, 8, 192)    331776      ['activation_74[0][0]']          
                                                                                                  
 batch_normalization_228 (Batch  (None, 8, 8, 320)   960         ['conv2d_270[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_232 (Batch  (None, 8, 8, 192)   576         ['conv2d_274[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_71 (Activation)     (None, 8, 8, 320)    0           ['batch_normalization_228[0][0]']
                                                                                                  
 activation_75 (Activation)     (None, 8, 8, 192)    0           ['batch_normalization_232[0][0]']
                                                                                                  
 max_pooling2d_84 (MaxPooling2D  (None, 8, 8, 768)   0           ['mixed7[0][0]']                 
 )                                                                                                
                                                                                                  
 mixed8 (Concatenate)           (None, 8, 8, 1280)   0           ['activation_71[0][0]',          
                                                                  'activation_75[0][0]',          
                                                                  'max_pooling2d_84[0][0]']       
                                                                                                  
 conv2d_279 (Conv2D)            (None, 8, 8, 448)    573440      ['mixed8[0][0]']                 
                                                                                                  
 batch_normalization_237 (Batch  (None, 8, 8, 448)   1344        ['conv2d_279[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_80 (Activation)     (None, 8, 8, 448)    0           ['batch_normalization_237[0][0]']
                                                                                                  
 conv2d_276 (Conv2D)            (None, 8, 8, 384)    491520      ['mixed8[0][0]']                 
                                                                                                  
 conv2d_280 (Conv2D)            (None, 8, 8, 384)    1548288     ['activation_80[0][0]']          
                                                                                                  
 batch_normalization_234 (Batch  (None, 8, 8, 384)   1152        ['conv2d_276[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_238 (Batch  (None, 8, 8, 384)   1152        ['conv2d_280[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_77 (Activation)     (None, 8, 8, 384)    0           ['batch_normalization_234[0][0]']
                                                                                                  
 activation_81 (Activation)     (None, 8, 8, 384)    0           ['batch_normalization_238[0][0]']
                                                                                                  
 conv2d_277 (Conv2D)            (None, 8, 8, 384)    442368      ['activation_77[0][0]']          
                                                                                                  
 conv2d_278 (Conv2D)            (None, 8, 8, 384)    442368      ['activation_77[0][0]']          
                                                                                                  
 conv2d_281 (Conv2D)            (None, 8, 8, 384)    442368      ['activation_81[0][0]']          
                                                                                                  
 conv2d_282 (Conv2D)            (None, 8, 8, 384)    442368      ['activation_81[0][0]']          
                                                                                                  
 average_pooling2d_14 (AverageP  (None, 8, 8, 1280)  0           ['mixed8[0][0]']                 
 ooling2D)                                                                                        
                                                                                                  
 conv2d_275 (Conv2D)            (None, 8, 8, 320)    409600      ['mixed8[0][0]']                 
                                                                                                  
 batch_normalization_235 (Batch  (None, 8, 8, 384)   1152        ['conv2d_277[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_236 (Batch  (None, 8, 8, 384)   1152        ['conv2d_278[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_239 (Batch  (None, 8, 8, 384)   1152        ['conv2d_281[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_240 (Batch  (None, 8, 8, 384)   1152        ['conv2d_282[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 conv2d_283 (Conv2D)            (None, 8, 8, 192)    245760      ['average_pooling2d_14[0][0]']   
                                                                                                  
 batch_normalization_233 (Batch  (None, 8, 8, 320)   960         ['conv2d_275[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_78 (Activation)     (None, 8, 8, 384)    0           ['batch_normalization_235[0][0]']
                                                                                                  
 activation_79 (Activation)     (None, 8, 8, 384)    0           ['batch_normalization_236[0][0]']
                                                                                                  
 activation_82 (Activation)     (None, 8, 8, 384)    0           ['batch_normalization_239[0][0]']
                                                                                                  
 activation_83 (Activation)     (None, 8, 8, 384)    0           ['batch_normalization_240[0][0]']
                                                                                                  
 batch_normalization_241 (Batch  (None, 8, 8, 192)   576         ['conv2d_283[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_76 (Activation)     (None, 8, 8, 320)    0           ['batch_normalization_233[0][0]']
                                                                                                  
 mixed9_0 (Concatenate)         (None, 8, 8, 768)    0           ['activation_78[0][0]',          
                                                                  'activation_79[0][0]']          
                                                                                                  
 concatenate (Concatenate)      (None, 8, 8, 768)    0           ['activation_82[0][0]',          
                                                                  'activation_83[0][0]']          
                                                                                                  
 activation_84 (Activation)     (None, 8, 8, 192)    0           ['batch_normalization_241[0][0]']
                                                                                                  
 mixed9 (Concatenate)           (None, 8, 8, 2048)   0           ['activation_76[0][0]',          
                                                                  'mixed9_0[0][0]',               
                                                                  'concatenate[0][0]',            
                                                                  'activation_84[0][0]']          
                                                                                                  
 conv2d_288 (Conv2D)            (None, 8, 8, 448)    917504      ['mixed9[0][0]']                 
                                                                                                  
 batch_normalization_246 (Batch  (None, 8, 8, 448)   1344        ['conv2d_288[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_89 (Activation)     (None, 8, 8, 448)    0           ['batch_normalization_246[0][0]']
                                                                                                  
 conv2d_285 (Conv2D)            (None, 8, 8, 384)    786432      ['mixed9[0][0]']                 
                                                                                                  
 conv2d_289 (Conv2D)            (None, 8, 8, 384)    1548288     ['activation_89[0][0]']          
                                                                                                  
 batch_normalization_243 (Batch  (None, 8, 8, 384)   1152        ['conv2d_285[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_247 (Batch  (None, 8, 8, 384)   1152        ['conv2d_289[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_86 (Activation)     (None, 8, 8, 384)    0           ['batch_normalization_243[0][0]']
                                                                                                  
 activation_90 (Activation)     (None, 8, 8, 384)    0           ['batch_normalization_247[0][0]']
                                                                                                  
 conv2d_286 (Conv2D)            (None, 8, 8, 384)    442368      ['activation_86[0][0]']          
                                                                                                  
 conv2d_287 (Conv2D)            (None, 8, 8, 384)    442368      ['activation_86[0][0]']          
                                                                                                  
 conv2d_290 (Conv2D)            (None, 8, 8, 384)    442368      ['activation_90[0][0]']          
                                                                                                  
 conv2d_291 (Conv2D)            (None, 8, 8, 384)    442368      ['activation_90[0][0]']          
                                                                                                  
 average_pooling2d_15 (AverageP  (None, 8, 8, 2048)  0           ['mixed9[0][0]']                 
 ooling2D)                                                                                        
                                                                                                  
 conv2d_284 (Conv2D)            (None, 8, 8, 320)    655360      ['mixed9[0][0]']                 
                                                                                                  
 batch_normalization_244 (Batch  (None, 8, 8, 384)   1152        ['conv2d_286[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_245 (Batch  (None, 8, 8, 384)   1152        ['conv2d_287[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_248 (Batch  (None, 8, 8, 384)   1152        ['conv2d_290[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 batch_normalization_249 (Batch  (None, 8, 8, 384)   1152        ['conv2d_291[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 conv2d_292 (Conv2D)            (None, 8, 8, 192)    393216      ['average_pooling2d_15[0][0]']   
                                                                                                  
 batch_normalization_242 (Batch  (None, 8, 8, 320)   960         ['conv2d_284[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_87 (Activation)     (None, 8, 8, 384)    0           ['batch_normalization_244[0][0]']
                                                                                                  
 activation_88 (Activation)     (None, 8, 8, 384)    0           ['batch_normalization_245[0][0]']
                                                                                                  
 activation_91 (Activation)     (None, 8, 8, 384)    0           ['batch_normalization_248[0][0]']
                                                                                                  
 activation_92 (Activation)     (None, 8, 8, 384)    0           ['batch_normalization_249[0][0]']
                                                                                                  
 batch_normalization_250 (Batch  (None, 8, 8, 192)   576         ['conv2d_292[0][0]']             
 Normalization)                                                                                   
                                                                                                  
 activation_85 (Activation)     (None, 8, 8, 320)    0           ['batch_normalization_242[0][0]']
                                                                                                  
 mixed9_1 (Concatenate)         (None, 8, 8, 768)    0           ['activation_87[0][0]',          
                                                                  'activation_88[0][0]']          
                                                                                                  
 concatenate_1 (Concatenate)    (None, 8, 8, 768)    0           ['activation_91[0][0]',          
                                                                  'activation_92[0][0]']          
                                                                                                  
 activation_93 (Activation)     (None, 8, 8, 192)    0           ['batch_normalization_250[0][0]']
                                                                                                  
 mixed10 (Concatenate)          (None, 8, 8, 2048)   0           ['activation_85[0][0]',          
                                                                  'mixed9_1[0][0]',               
                                                                  'concatenate_1[0][0]',          
                                                                  'activation_93[0][0]']          
                                                                                                  
 avg_pool (GlobalAveragePooling  (None, 2048)        0           ['mixed10[0][0]']                
 2D)                                                                                              
                                                                                                  
 predictions (Dense)            (None, 1000)         2049000     ['avg_pool[0][0]']               
                                                                                                  
==================================================================================================
Total params: 23,851,784
Trainable params: 23,817,352
Non-trainable params: 34,432
__________________________________________________________________________________________________

The above summary tells us about the format of InceptionNet v3, showing how it uses BatchNormalization and Activation layers after each convolutional layer. We shall take note of this, and apply it later on. In addition, we shall also simplify the InceptionNet model, given that it was optimized for 299 x 299 images, and not the 128 x 128 and 31 x 31 images of our dataset.

---

Initial Modelling (31 x 31)

---

def vgg_block(n_conv, n_filter, pooling = True):
    block = Sequential()
    for _ in range (n_conv):
        block.add(Conv2D(n_filter, kernel_size=3, padding='same', activation='relu'))
        block.add(BatchNormalization())
        block.add(ReLU())
    if pooling:
        block.add(MaxPooling2D(pool_size=2, strides=2))
    return block

Here, we make a simpler model based on the VGG architecture. We make use of 10 Convolutional Layers, and 3 Fully Connected layers, making it a VGG-13 model. In addition, instead of using a 1000 channel softmax layer, we use a 15 channel softmax layer to match the output of our dataset.

res31_vggmodel = Sequential()
res31_vggmodel.add(Input(shape = (31, 31, 1)))
res31_vggmodel.add(vgg_block(2, 32))
res31_vggmodel.add(vgg_block(2, 64))
res31_vggmodel.add(vgg_block(3, 128))
res31_vggmodel.add(vgg_block(3, 256))
res31_vggmodel.add(Flatten())
res31_vggmodel.add(Dense(4096, activation = "relu"))
res31_vggmodel.add(Dense(4096, activation = "relu"))
res31_vggmodel.add(ActivityRegularization(l1 = 0.0001, l2 = 0.0001))
res31_vggmodel.add(Dense(num_classes, activation = "softmax"))
res31_vggmodel.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])

history = res31_vggmodel.fit(std_res31_train, validation_data = std_res31_validation, epochs = 150, callbacks = earlyStopper)
cnnf.model_metric_graph(history, res31_vggmodel, std_res31_validation, name = "adapted VGG-13", class_names = class_names)

Epoch 1/150
283/283 [==============================] - 9s 25ms/step - loss: 2.4111 - accuracy: 0.2172 - val_loss: 2.9013 - val_accuracy: 0.0813
Epoch 2/150
283/283 [==============================] - 7s 25ms/step - loss: 1.7566 - accuracy: 0.4350 - val_loss: 1.9467 - val_accuracy: 0.3463
Epoch 3/150
283/283 [==============================] - 7s 24ms/step - loss: 1.3811 - accuracy: 0.5624 - val_loss: 1.5058 - val_accuracy: 0.5510
Epoch 4/150
283/283 [==============================] - 8s 26ms/step - loss: 1.0904 - accuracy: 0.6615 - val_loss: 1.4079 - val_accuracy: 0.5523
Epoch 5/150
283/283 [==============================] - 7s 25ms/step - loss: 0.8906 - accuracy: 0.7370 - val_loss: 1.1108 - val_accuracy: 0.6773
Epoch 6/150
283/283 [==============================] - 7s 24ms/step - loss: 0.7268 - accuracy: 0.7836 - val_loss: 1.1410 - val_accuracy: 0.6817
Epoch 7/150
283/283 [==============================] - 8s 25ms/step - loss: 0.5670 - accuracy: 0.8365 - val_loss: 0.7813 - val_accuracy: 0.7803
Epoch 8/150
283/283 [==============================] - 7s 25ms/step - loss: 0.4718 - accuracy: 0.8658 - val_loss: 0.7053 - val_accuracy: 0.8083
Epoch 9/150
283/283 [==============================] - 7s 25ms/step - loss: 0.3791 - accuracy: 0.8916 - val_loss: 1.0165 - val_accuracy: 0.7303
Epoch 10/150
283/283 [==============================] - 7s 24ms/step - loss: 0.3587 - accuracy: 0.8998 - val_loss: 0.6048 - val_accuracy: 0.8360
Epoch 11/150
283/283 [==============================] - 7s 24ms/step - loss: 0.3008 - accuracy: 0.9208 - val_loss: 0.6989 - val_accuracy: 0.8320
Epoch 12/150
283/283 [==============================] - 7s 25ms/step - loss: 0.2578 - accuracy: 0.9302 - val_loss: 0.7067 - val_accuracy: 0.8177
Epoch 13/150
283/283 [==============================] - 7s 24ms/step - loss: 0.2345 - accuracy: 0.9343 - val_loss: 0.6134 - val_accuracy: 0.8400
Epoch 14/150
283/283 [==============================] - 7s 24ms/step - loss: 0.1990 - accuracy: 0.9467 - val_loss: 0.8285 - val_accuracy: 0.8103
Epoch 15/150
283/283 [==============================] - 7s 24ms/step - loss: 0.1759 - accuracy: 0.9548 - val_loss: 0.5408 - val_accuracy: 0.8577
Epoch 16/150
283/283 [==============================] - 9s 29ms/step - loss: 0.1414 - accuracy: 0.9620 - val_loss: 0.5555 - val_accuracy: 0.8747
Epoch 17/150
283/283 [==============================] - 8s 28ms/step - loss: 0.1526 - accuracy: 0.9597 - val_loss: 0.5036 - val_accuracy: 0.8693
Epoch 18/150
283/283 [==============================] - 8s 27ms/step - loss: 0.1705 - accuracy: 0.9554 - val_loss: 0.4093 - val_accuracy: 0.8953
Epoch 19/150
283/283 [==============================] - 8s 25ms/step - loss: 0.1204 - accuracy: 0.9704 - val_loss: 0.3586 - val_accuracy: 0.9217
Epoch 20/150
283/283 [==============================] - 8s 26ms/step - loss: 0.1165 - accuracy: 0.9684 - val_loss: 0.5269 - val_accuracy: 0.8737
Epoch 21/150
283/283 [==============================] - 8s 26ms/step - loss: 0.1183 - accuracy: 0.9710 - val_loss: 0.5190 - val_accuracy: 0.8837
Epoch 22/150
283/283 [==============================] - 8s 29ms/step - loss: 0.1477 - accuracy: 0.9626 - val_loss: 0.4237 - val_accuracy: 0.8927
Epoch 23/150
283/283 [==============================] - 8s 26ms/step - loss: 0.1293 - accuracy: 0.9671 - val_loss: 0.4876 - val_accuracy: 0.8903
Epoch 24/150
283/283 [==============================] - 8s 27ms/step - loss: 0.1560 - accuracy: 0.9592 - val_loss: 0.4937 - val_accuracy: 0.8860
Epoch 25/150
283/283 [==============================] - 8s 28ms/step - loss: 0.1336 - accuracy: 0.9668 - val_loss: 0.4204 - val_accuracy: 0.8923
Epoch 26/150
283/283 [==============================] - 7s 25ms/step - loss: 0.0763 - accuracy: 0.9818 - val_loss: 0.3732 - val_accuracy: 0.9110
Epoch 27/150
283/283 [==============================] - 8s 26ms/step - loss: 0.0749 - accuracy: 0.9825 - val_loss: 0.4512 - val_accuracy: 0.8963
Epoch 28/150
283/283 [==============================] - 7s 25ms/step - loss: 0.0717 - accuracy: 0.9828 - val_loss: 0.4216 - val_accuracy: 0.9067
Epoch 29/150
283/283 [==============================] - 7s 25ms/step - loss: 0.0731 - accuracy: 0.9811 - val_loss: 0.4964 - val_accuracy: 0.8863
Epoch 30/150
283/283 [==============================] - 7s 25ms/step - loss: 0.1283 - accuracy: 0.9668 - val_loss: 0.5020 - val_accuracy: 0.8883
Epoch 31/150
283/283 [==============================] - 7s 25ms/step - loss: 0.0681 - accuracy: 0.9838 - val_loss: 0.4358 - val_accuracy: 0.9120
Epoch 32/150
283/283 [==============================] - 7s 24ms/step - loss: 0.1151 - accuracy: 0.9711 - val_loss: 0.5898 - val_accuracy: 0.8790
Epoch 33/150
283/283 [==============================] - 8s 26ms/step - loss: 0.0729 - accuracy: 0.9829 - val_loss: 0.4331 - val_accuracy: 0.9047
Epoch 34/150
283/283 [==============================] - 8s 26ms/step - loss: 0.0736 - accuracy: 0.9824 - val_loss: 0.4451 - val_accuracy: 0.9130
Epoch 35/150
283/283 [==============================] - 7s 24ms/step - loss: 0.0541 - accuracy: 0.9875 - val_loss: 0.5297 - val_accuracy: 0.8847
Epoch 36/150
283/283 [==============================] - 7s 25ms/step - loss: 0.0573 - accuracy: 0.9865 - val_loss: 0.6814 - val_accuracy: 0.8583
Epoch 37/150
283/283 [==============================] - 7s 25ms/step - loss: 0.0711 - accuracy: 0.9855 - val_loss: 0.3943 - val_accuracy: 0.9147
Epoch 38/150
283/283 [==============================] - 7s 24ms/step - loss: 0.0511 - accuracy: 0.9888 - val_loss: 0.3811 - val_accuracy: 0.9183
Epoch 39/150
283/283 [==============================] - 7s 25ms/step - loss: 0.0931 - accuracy: 0.9763 - val_loss: 0.4722 - val_accuracy: 0.8983
Epoch 40/150
283/283 [==============================] - 8s 29ms/step - loss: 0.0804 - accuracy: 0.9804 - val_loss: 0.4319 - val_accuracy: 0.9040
Epoch 41/150
283/283 [==============================] - 8s 27ms/step - loss: 0.0623 - accuracy: 0.9862 - val_loss: 0.4914 - val_accuracy: 0.8903
Epoch 42/150
283/283 [==============================] - 8s 27ms/step - loss: 0.0852 - accuracy: 0.9802 - val_loss: 0.4446 - val_accuracy: 0.9070
Epoch 43/150
283/283 [==============================] - 8s 27ms/step - loss: 0.0448 - accuracy: 0.9906 - val_loss: 0.4817 - val_accuracy: 0.9033
Epoch 44/150
283/283 [==============================] - 7s 25ms/step - loss: 0.0564 - accuracy: 0.9872 - val_loss: 0.4102 - val_accuracy: 0.9153
Epoch 45/150
283/283 [==============================] - 8s 26ms/step - loss: 0.1262 - accuracy: 0.9679 - val_loss: 0.4463 - val_accuracy: 0.8913
Epoch 46/150
283/283 [==============================] - 8s 26ms/step - loss: 0.0392 - accuracy: 0.9898 - val_loss: 0.3822 - val_accuracy: 0.9283
Epoch 47/150
283/283 [==============================] - 8s 26ms/step - loss: 0.0274 - accuracy: 0.9946 - val_loss: 0.5049 - val_accuracy: 0.9100
Epoch 48/150
283/283 [==============================] - 7s 24ms/step - loss: 0.0448 - accuracy: 0.9900 - val_loss: 0.4839 - val_accuracy: 0.8983
Epoch 49/150
283/283 [==============================] - 7s 24ms/step - loss: 0.0579 - accuracy: 0.9860 - val_loss: 0.4841 - val_accuracy: 0.8953

res31_vggmodel.summary()

Model: "sequential_90"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 sequential_91 (Sequential)  (None, 15, 15, 32)        9824      
                                                                 
 sequential_92 (Sequential)  (None, 7, 7, 64)          55936     
                                                                 
 sequential_93 (Sequential)  (None, 3, 3, 128)         370560    
                                                                 
 sequential_94 (Sequential)  (None, 1, 1, 256)         1478400   
                                                                 
 flatten_19 (Flatten)        (None, 256)               0         
                                                                 
 dense_55 (Dense)            (None, 4096)              1052672   
                                                                 
 dense_56 (Dense)            (None, 4096)              16781312  
                                                                 
 activity_regularization_19   (None, 4096)             0         
 (ActivityRegularization)                                        
                                                                 
 dense_57 (Dense)            (None, 15)                61455     
                                                                 
=================================================================
Total params: 19,810,159
Trainable params: 19,807,471
Non-trainable params: 2,688
_________________________________________________________________

tf.keras.utils.plot_model(res31_vggmodel, show_shapes=True, dpi=80)

From the above, we can see that the VGG architecture performs quite well on our dataset. Now we shall try to create a InceptionNet model, and define the Inception Modules as well.

Defining "Inception Modules" for use later.

"inceptionBlock1" follows this architecture:

class inceptionBlock1(Layer):
    def __init__(self, num_filters, **kwargs):
        super(inceptionBlock1, self).__init__(**kwargs)
        self.num_filters = num_filters

    def build(self, input_shape):
        self.tower_1_conv1 = Conv2D(self.num_filters, (1, 1), padding='same', activation='relu')
        self.tower_1_conv2 = Conv2D(self.num_filters, (3, 3), padding='same', activation='relu')
        self.tower_1_conv3 = Conv2D(self.num_filters, (3, 3), padding='same', activation='relu')

        self.tower_2_conv1 = Conv2D(self.num_filters, (1, 1), padding='same', activation='relu')
        self.tower_2_conv2 = Conv2D(self.num_filters, (3, 3), padding='same', activation='relu')

        self.tower_3_pool = AveragePooling2D((3, 3), strides=(1, 1), padding='same')
        self.tower_3_conv1 = Conv2D(self.num_filters, (1, 1), padding='same', activation='relu')

        self.tower_4_conv1 = Conv2D(self.num_filters, (1, 1), padding='same', activation='relu')

    def call(self, inputs):
        tower_1 = self.tower_1_conv1(inputs)
        tower_1 = self.tower_1_conv2(tower_1)
        tower_1 = self.tower_1_conv3(tower_1)

        tower_2 = self.tower_2_conv1(inputs)
        tower_2 = self.tower_2_conv2(tower_2)

        tower_3 = self.tower_3_pool(inputs)
        tower_3 = self.tower_3_conv1(tower_3)

        tower_4 = self.tower_4_conv1(inputs)

        output = concatenate([tower_1, tower_2, tower_3, tower_4], axis=3)
        return output

"inceptionBlock2" follows this architecture:

note: for our use, n shall be set at 3, representing a 3 x 3 factorized convolution.

class inceptionBlock2(Layer):
    def __init__(self, num_filters, n, **kwargs):
        super(inceptionBlock2, self).__init__(**kwargs)
        self.num_filters = num_filters
        self.n = n

    def build(self, input_shape):
        self.tower_1_conv1 = Conv2D(self.num_filters, (1, 1), padding='same', activation='relu')
        self.tower_1_conv2 = Conv2D(self.num_filters, (1, self.n), padding='same', activation='relu')
        self.tower_1_conv3 = Conv2D(self.num_filters, (self.n, 1), padding='same', activation='relu')
        self.tower_1_conv4 = Conv2D(self.num_filters, (1, self.n), padding='same', activation='relu')
        self.tower_1_conv5 = Conv2D(self.num_filters, (self.n, 1), padding='same', activation='relu')

        self.tower_2_conv1 = Conv2D(self.num_filters, (1, 1), padding='same', activation='relu')
        self.tower_2_conv2 = Conv2D(self.num_filters, (1, self.n), padding='same', activation='relu')
        self.tower_2_conv3 = Conv2D(self.num_filters, (self.n, 1), padding='same', activation='relu')

        self.tower_3_pool = AveragePooling2D((3, 3), strides=(1, 1), padding='same')
        self.tower_3_conv1 = Conv2D(self.num_filters, (1, 1), padding='same', activation='relu')

        self.tower_4_conv1 = Conv2D(self.num_filters, (1, 1), padding='same', activation='relu')


    def call(self, inputs):
        tower_1 = self.tower_1_conv1(inputs)
        tower_1 = self.tower_1_conv2(tower_1)
        tower_1 = self.tower_1_conv3(tower_1)
        tower_1 = self.tower_1_conv4(tower_1)
        tower_1 = self.tower_1_conv5(tower_1)

        tower_2 = self.tower_2_conv1(inputs)
        tower_2 = self.tower_2_conv2(tower_2)
        tower_2 = self.tower_2_conv3(tower_2)

        tower_3 = self.tower_3_pool(inputs)
        tower_3 = self.tower_3_conv1(tower_3)

        tower_4 = self.tower_4_conv1(inputs)

        output = concatenate([tower_1, tower_2, tower_3, tower_4], axis=3)
        return output

"inceptionBlock3" follows this architecture:

class inceptionBlock3(Layer):
    def __init__(self, num_filters, **kwargs):
        super(inceptionBlock3, self).__init__(**kwargs)
        self.num_filters = num_filters

    def build(self, input_shape):
        self.branch1_conv1 = Conv2D(self.num_filters, (1, 1), padding='same', activation='relu')

        self.branch2_conv1 = Conv2D(self.num_filters, (1, 1), padding='same', activation='relu')
        self.branch2_conv2 = Conv2D(self.num_filters, (1, 3), padding='same', activation='relu')
        self.branch2_conv3 = Conv2D(self.num_filters, (3, 1), padding='same', activation='relu')

        self.branch3_conv1 = Conv2D(self.num_filters, (1, 1), padding='same', activation='relu')
        self.branch3_conv2 = Conv2D(self.num_filters, (3, 3), padding='same', activation='relu')
        self.branch3_conv3 = Conv2D(self.num_filters, (1, 3), padding='same', activation='relu')
        self.branch3_conv4 = Conv2D(self.num_filters, (3, 1), padding='same', activation='relu')

        self.branch4_pool = AveragePooling2D((3, 3), strides=(1, 1), padding='same')
        self.branch4_conv1 = Conv2D(self.num_filters, (1, 1), padding='same', activation='relu')

    def call(self, inputs):
        branch1 = self.branch1_conv1(inputs)

        branch2 = self.branch2_conv1(inputs)
        branch2_1x3 = self.branch2_conv2(branch2)
        branch2_3x1 = self.branch2_conv3(branch2)

        branch3 = self.branch3_conv1(inputs)
        branch3 = self.branch3_conv2(branch3)
        branch3_1x3 = self.branch3_conv3(branch3)
        branch3_3x1 = self.branch3_conv4(branch3)

        branch4 = self.branch4_pool(inputs)
        branch4 = self.branch4_conv1(branch4)

        output = concatenate([branch1, branch2_1x3, branch2_3x1, branch3_1x3, branch3_3x1, branch4], axis=-1)
        return output

Here, we define a simpler version of the InceptionNet v3 model. Instead of using a inception block layout of 3-5-2, we shall use a simple layour of 2-3-1. We also remove two more convolutional layers to simplify it further, but add a regularization layer to stabilize the model.

res31_inceptionmodel = Sequential()
res31_inceptionmodel.add(Conv2D(64, kernel_size = (3,3), strides = (2,1), padding = "same", activation = "relu"))
res31_inceptionmodel.add(Conv2D(64, kernel_size = (3,3), strides = (1,1), padding = "valid", activation = "relu"))
res31_inceptionmodel.add(MaxPooling2D((2,2)))
res31_inceptionmodel.add(Conv2D(64, kernel_size = (3,3), strides = (1,1), padding = "same", activation = "relu"))
res31_inceptionmodel.add(Conv2D(64, kernel_size = (3,3), strides = (1,1), padding = "same", activation = "relu"))
res31_inceptionmodel.add(inceptionBlock1(64))
res31_inceptionmodel.add(inceptionBlock1(64))
res31_inceptionmodel.add(inceptionBlock2(num_filters = 64, n = 3))
res31_inceptionmodel.add(inceptionBlock2(num_filters = 64, n = 3))
res31_inceptionmodel.add(inceptionBlock2(num_filters = 64, n = 3))
res31_inceptionmodel.add(inceptionBlock3(num_filters = 64))
res31_inceptionmodel.add(AveragePooling2D())
res31_inceptionmodel.add(Dropout(0.2))
res31_inceptionmodel.add(Dense(2048, activation = "relu"))
res31_inceptionmodel.add(ActivityRegularization(l1 = 0.0005, l2 = 0.0002))
res31_inceptionmodel.add(Flatten()) #added to allow model to work 
res31_inceptionmodel.add(Dense(num_classes, activation = "softmax"))
res31_inceptionmodel.compile(optimizer='rmsprop', loss='sparse_categorical_crossentropy', metrics=['accuracy'])

history = res31_inceptionmodel.fit(std_res31_train, validation_data = std_res31_validation, epochs = 150, callbacks = earlyStopper)
cnnf.model_metric_graph(history, res31_inceptionmodel, std_res31_validation, name = "adapted InceptionNet v3", class_names = class_names)

Epoch 1/150
283/283 [==============================] - 20s 49ms/step - loss: 2.5695 - accuracy: 0.1500 - val_loss: 2.5545 - val_accuracy: 0.1227
Epoch 2/150
283/283 [==============================] - 13s 46ms/step - loss: 2.1079 - accuracy: 0.3210 - val_loss: 2.1597 - val_accuracy: 0.3043
Epoch 3/150
283/283 [==============================] - 13s 47ms/step - loss: 1.7902 - accuracy: 0.4403 - val_loss: 3.3991 - val_accuracy: 0.2017
Epoch 4/150
283/283 [==============================] - 14s 48ms/step - loss: 1.5317 - accuracy: 0.5246 - val_loss: 2.6475 - val_accuracy: 0.3013
Epoch 5/150
283/283 [==============================] - 14s 48ms/step - loss: 1.3162 - accuracy: 0.5961 - val_loss: 1.2645 - val_accuracy: 0.6080
Epoch 6/150
283/283 [==============================] - 13s 46ms/step - loss: 1.1083 - accuracy: 0.6577 - val_loss: 1.2395 - val_accuracy: 0.6470
Epoch 7/150
283/283 [==============================] - 13s 46ms/step - loss: 0.9549 - accuracy: 0.7119 - val_loss: 2.3906 - val_accuracy: 0.5113
Epoch 8/150
283/283 [==============================] - 13s 46ms/step - loss: 0.8167 - accuracy: 0.7622 - val_loss: 1.2736 - val_accuracy: 0.6290
Epoch 9/150
283/283 [==============================] - 14s 48ms/step - loss: 0.6860 - accuracy: 0.7987 - val_loss: 2.2882 - val_accuracy: 0.5150
Epoch 10/150
283/283 [==============================] - 14s 48ms/step - loss: 0.5921 - accuracy: 0.8292 - val_loss: 0.9844 - val_accuracy: 0.7477
Epoch 11/150
283/283 [==============================] - 13s 46ms/step - loss: 0.5078 - accuracy: 0.8571 - val_loss: 1.0955 - val_accuracy: 0.6730
Epoch 12/150
283/283 [==============================] - 13s 45ms/step - loss: 0.4403 - accuracy: 0.8768 - val_loss: 0.9864 - val_accuracy: 0.7353
Epoch 13/150
283/283 [==============================] - 14s 48ms/step - loss: 0.3940 - accuracy: 0.8949 - val_loss: 1.4728 - val_accuracy: 0.6500
Epoch 14/150
283/283 [==============================] - 14s 48ms/step - loss: 0.3401 - accuracy: 0.9122 - val_loss: 0.9465 - val_accuracy: 0.7473
Epoch 15/150
283/283 [==============================] - 14s 47ms/step - loss: 0.2990 - accuracy: 0.9243 - val_loss: 1.2581 - val_accuracy: 0.7013
Epoch 16/150
283/283 [==============================] - 13s 46ms/step - loss: 0.2790 - accuracy: 0.9279 - val_loss: 0.7852 - val_accuracy: 0.8050
Epoch 17/150
283/283 [==============================] - 13s 46ms/step - loss: 0.2454 - accuracy: 0.9406 - val_loss: 0.7924 - val_accuracy: 0.8127
Epoch 18/150
283/283 [==============================] - 14s 48ms/step - loss: 0.2421 - accuracy: 0.9427 - val_loss: 0.7448 - val_accuracy: 0.8257
Epoch 19/150
283/283 [==============================] - 14s 48ms/step - loss: 0.2201 - accuracy: 0.9489 - val_loss: 1.1154 - val_accuracy: 0.7400
Epoch 20/150
283/283 [==============================] - 14s 48ms/step - loss: 0.2030 - accuracy: 0.9555 - val_loss: 0.7546 - val_accuracy: 0.8320
Epoch 21/150
283/283 [==============================] - 13s 47ms/step - loss: 0.1911 - accuracy: 0.9580 - val_loss: 0.8820 - val_accuracy: 0.7953
Epoch 22/150
283/283 [==============================] - 14s 47ms/step - loss: 0.1837 - accuracy: 0.9579 - val_loss: 1.5478 - val_accuracy: 0.6967
Epoch 23/150
283/283 [==============================] - 14s 47ms/step - loss: 0.1799 - accuracy: 0.9581 - val_loss: 0.7243 - val_accuracy: 0.8307
Epoch 24/150
283/283 [==============================] - 14s 48ms/step - loss: 0.1578 - accuracy: 0.9653 - val_loss: 0.7525 - val_accuracy: 0.8267
Epoch 25/150
283/283 [==============================] - 13s 46ms/step - loss: 0.1518 - accuracy: 0.9687 - val_loss: 0.9391 - val_accuracy: 0.8047
Epoch 26/150
283/283 [==============================] - 14s 47ms/step - loss: 0.1516 - accuracy: 0.9675 - val_loss: 0.8988 - val_accuracy: 0.7977
Epoch 27/150
283/283 [==============================] - 14s 47ms/step - loss: 0.1376 - accuracy: 0.9713 - val_loss: 0.6745 - val_accuracy: 0.8440
Epoch 28/150
283/283 [==============================] - 13s 46ms/step - loss: 0.1327 - accuracy: 0.9715 - val_loss: 0.7117 - val_accuracy: 0.8397
Epoch 29/150
283/283 [==============================] - 14s 47ms/step - loss: 0.1310 - accuracy: 0.9713 - val_loss: 0.7146 - val_accuracy: 0.8437
Epoch 30/150
283/283 [==============================] - 13s 46ms/step - loss: 0.1267 - accuracy: 0.9735 - val_loss: 0.7260 - val_accuracy: 0.8400
Epoch 31/150
283/283 [==============================] - 13s 46ms/step - loss: 0.1163 - accuracy: 0.9780 - val_loss: 0.8078 - val_accuracy: 0.8183
Epoch 32/150
283/283 [==============================] - 13s 46ms/step - loss: 0.1270 - accuracy: 0.9747 - val_loss: 1.0474 - val_accuracy: 0.8053
Epoch 33/150
283/283 [==============================] - 13s 46ms/step - loss: 0.1192 - accuracy: 0.9743 - val_loss: 0.6132 - val_accuracy: 0.8587
Epoch 34/150
283/283 [==============================] - 14s 47ms/step - loss: 0.1082 - accuracy: 0.9783 - val_loss: 0.8247 - val_accuracy: 0.8193
Epoch 35/150
283/283 [==============================] - 14s 48ms/step - loss: 0.1112 - accuracy: 0.9778 - val_loss: 0.8026 - val_accuracy: 0.8257
Epoch 36/150
283/283 [==============================] - 13s 46ms/step - loss: 0.1132 - accuracy: 0.9772 - val_loss: 0.6939 - val_accuracy: 0.8470
Epoch 37/150
283/283 [==============================] - 14s 47ms/step - loss: 0.1097 - accuracy: 0.9781 - val_loss: 0.6901 - val_accuracy: 0.8417
Epoch 38/150
283/283 [==============================] - 14s 48ms/step - loss: 0.1064 - accuracy: 0.9791 - val_loss: 0.7908 - val_accuracy: 0.8353
Epoch 39/150
283/283 [==============================] - 13s 46ms/step - loss: 0.0979 - accuracy: 0.9804 - val_loss: 0.6912 - val_accuracy: 0.8513
Epoch 40/150
283/283 [==============================] - 13s 46ms/step - loss: 0.1076 - accuracy: 0.9790 - val_loss: 0.7810 - val_accuracy: 0.8340
Epoch 41/150
283/283 [==============================] - 14s 47ms/step - loss: 0.1018 - accuracy: 0.9786 - val_loss: 0.7506 - val_accuracy: 0.8447
Epoch 42/150
283/283 [==============================] - 14s 48ms/step - loss: 0.0938 - accuracy: 0.9817 - val_loss: 0.7034 - val_accuracy: 0.8517
Epoch 43/150
283/283 [==============================] - 14s 48ms/step - loss: 0.0985 - accuracy: 0.9801 - val_loss: 0.7050 - val_accuracy: 0.8477
Epoch 44/150
283/283 [==============================] - 13s 46ms/step - loss: 0.0970 - accuracy: 0.9797 - val_loss: 0.6531 - val_accuracy: 0.8567
Epoch 45/150
283/283 [==============================] - 13s 47ms/step - loss: 0.0978 - accuracy: 0.9774 - val_loss: 0.7542 - val_accuracy: 0.8450
Epoch 46/150
283/283 [==============================] - 14s 47ms/step - loss: 0.0914 - accuracy: 0.9829 - val_loss: 1.1763 - val_accuracy: 0.7930
Epoch 47/150
283/283 [==============================] - 14s 47ms/step - loss: 0.0888 - accuracy: 0.9823 - val_loss: 0.8132 - val_accuracy: 0.8477
Epoch 48/150
283/283 [==============================] - 14s 48ms/step - loss: 0.0952 - accuracy: 0.9815 - val_loss: 1.5317 - val_accuracy: 0.7497
Epoch 49/150
283/283 [==============================] - 13s 47ms/step - loss: 0.0927 - accuracy: 0.9808 - val_loss: 0.7616 - val_accuracy: 0.8283
Epoch 50/150
283/283 [==============================] - 14s 47ms/step - loss: 0.0840 - accuracy: 0.9823 - val_loss: 0.7469 - val_accuracy: 0.8513
Epoch 51/150
283/283 [==============================] - 14s 47ms/step - loss: 0.0951 - accuracy: 0.9812 - val_loss: 0.8766 - val_accuracy: 0.8267
Epoch 52/150
283/283 [==============================] - 14s 48ms/step - loss: 0.0906 - accuracy: 0.9806 - val_loss: 0.6838 - val_accuracy: 0.8580
Epoch 53/150
283/283 [==============================] - 13s 47ms/step - loss: 0.0909 - accuracy: 0.9806 - val_loss: 0.7922 - val_accuracy: 0.8383
Epoch 54/150
283/283 [==============================] - 13s 46ms/step - loss: 0.0920 - accuracy: 0.9812 - val_loss: 0.7282 - val_accuracy: 0.8497
Epoch 55/150
283/283 [==============================] - 14s 47ms/step - loss: 0.0856 - accuracy: 0.9842 - val_loss: 1.9478 - val_accuracy: 0.6517
Epoch 56/150
283/283 [==============================] - 14s 47ms/step - loss: 0.0960 - accuracy: 0.9811 - val_loss: 0.7762 - val_accuracy: 0.8300
Epoch 57/150
283/283 [==============================] - 14s 47ms/step - loss: 0.0867 - accuracy: 0.9822 - val_loss: 1.0144 - val_accuracy: 0.8083
Epoch 58/150
283/283 [==============================] - 14s 47ms/step - loss: 0.0893 - accuracy: 0.9843 - val_loss: 0.6928 - val_accuracy: 0.8500
Epoch 59/150
283/283 [==============================] - 13s 46ms/step - loss: 0.0795 - accuracy: 0.9833 - val_loss: 0.7947 - val_accuracy: 0.8407
Epoch 60/150
283/283 [==============================] - 14s 47ms/step - loss: 0.0825 - accuracy: 0.9842 - val_loss: 0.6766 - val_accuracy: 0.8613
Epoch 61/150
283/283 [==============================] - 14s 48ms/step - loss: 0.0918 - accuracy: 0.9818 - val_loss: 0.7583 - val_accuracy: 0.8503
Epoch 62/150
283/283 [==============================] - 13s 45ms/step - loss: 0.0834 - accuracy: 0.9813 - val_loss: 0.7093 - val_accuracy: 0.8480
Epoch 63/150
283/283 [==============================] - 13s 45ms/step - loss: 0.0767 - accuracy: 0.9854 - val_loss: 0.7206 - val_accuracy: 0.8510

res31_inceptionmodel.summary()

Model: "sequential_95"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 conv2d_303 (Conv2D)         (None, 16, 31, 64)        640       
                                                                 
 conv2d_304 (Conv2D)         (None, 14, 29, 64)        36928     
                                                                 
 max_pooling2d_89 (MaxPoolin  (None, 7, 14, 64)        0         
 g2D)                                                            
                                                                 
 conv2d_305 (Conv2D)         (None, 7, 14, 64)         36928     
                                                                 
 conv2d_306 (Conv2D)         (None, 7, 14, 64)         36928     
                                                                 
 inception_block1_20 (incept  (None, 7, 14, 256)       127424    
 ionBlock1)                                                      
                                                                 
 inception_block1_21 (incept  (None, 7, 14, 256)       176576    
 ionBlock1)                                                      
                                                                 
 inception_block2_33 (incept  (None, 7, 14, 256)       139904    
 ionBlock2)                                                      
                                                                 
 inception_block2_34 (incept  (None, 7, 14, 256)       139904    
 ionBlock2)                                                      
                                                                 
 inception_block2_35 (incept  (None, 7, 14, 256)       139904    
 ionBlock2)                                                      
                                                                 
 inception_block3_13 (incept  (None, 7, 14, 384)       152128    
 ionBlock3)                                                      
                                                                 
 average_pooling2d_16 (Avera  (None, 3, 7, 384)        0         
 gePooling2D)                                                    
                                                                 
 dropout_7 (Dropout)         (None, 3, 7, 384)         0         
                                                                 
 dense_58 (Dense)            (None, 3, 7, 2048)        788480    
                                                                 
 activity_regularization_20   (None, 3, 7, 2048)       0         
 (ActivityRegularization)                                        
                                                                 
 flatten_20 (Flatten)        (None, 43008)             0         
                                                                 
 dense_59 (Dense)            (None, 15)                645135    
                                                                 
=================================================================
Total params: 2,420,879
Trainable params: 2,420,879
Non-trainable params: 0
_________________________________________________________________

tf.keras.utils.plot_model(res31_inceptionmodel, show_shapes=True, dpi=80)

finalres31_model_list = [res31_vggmodel, res31_inceptionmodel, baseline31luminosity_model]
finalres31_data_list = [std_res31_validation, std_res31_validation, res31luminosity_validation_data]
name_list = ["Modified VGG-13", "Modified Inception v3", "Baseline CNN"]
cnnf.model_comparison(finalres31_model_list, finalres31_data_list, name_list)

94/94 [==============================] - 1s 11ms/step - loss: 0.4841 - accuracy: 0.8953
94/94 [==============================] - 2s 14ms/step - loss: 0.7206 - accuracy: 0.8510
94/94 [==============================] - 1s 9ms/step - loss: 1.2177 - accuracy: 0.6267

From the above, we can see that the VGG-13 model performs the best, better than the baseline CNN model we used, and much better than the InceptionNet v3 model that we attempted to use. Hence, from this, we shall tune the VGG-13 model, and improve it for the final model.

---

Initial Modelling (128 x 128)

---

Here, we attempt to make a model for our data, based off of the VGG architecture. For this model, we make use of 13 convolutional layers, and 3 Fully Connected (Dense) layers, making it a VGG-16 model. In addition, instead of using a 1000 channel softmax layer, we use a 15 channel softmax layer to match the output of our dataset.

res128_vggmodel = Sequential()
res128_vggmodel.add(Input(shape = (128, 128, 1)))
res128_vggmodel.add(vgg_block(2, 32))
res128_vggmodel.add(vgg_block(2, 64))
res128_vggmodel.add(vgg_block(3, 128))
res128_vggmodel.add(vgg_block(3, 256))
res128_vggmodel.add(vgg_block(3, 256))
res128_vggmodel.add(Dense(4096, activation = "relu"))
res128_vggmodel.add(Dense(4096, activation = "relu"))
res128_vggmodel.add(ActivityRegularization(l1 = 0.00001, l2 = 0.00001))
res128_vggmodel.add(Flatten())
res128_vggmodel.add(Dense(num_classes, activation = "softmax"))
res128_vggmodel.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])

history = res128_vggmodel.fit(std_res128_train, validation_data = std_res128_validation, epochs = 150, callbacks = earlyStopper)
cnnf.model_metric_graph(history, res128_vggmodel, std_res128_validation, name = "adapted VGG-16", class_names = class_names)

Epoch 1/150
283/283 [==============================] - 31s 104ms/step - loss: 2.3772 - accuracy: 0.2537 - val_loss: 3.4935 - val_accuracy: 0.0663
Epoch 2/150
283/283 [==============================] - 29s 102ms/step - loss: 1.5532 - accuracy: 0.4900 - val_loss: 2.0382 - val_accuracy: 0.3340
Epoch 3/150
283/283 [==============================] - 29s 102ms/step - loss: 1.2367 - accuracy: 0.6079 - val_loss: 2.6715 - val_accuracy: 0.3583
Epoch 4/150
283/283 [==============================] - 29s 102ms/step - loss: 0.9627 - accuracy: 0.6995 - val_loss: 1.1481 - val_accuracy: 0.6597
Epoch 5/150
283/283 [==============================] - 29s 102ms/step - loss: 0.7828 - accuracy: 0.7605 - val_loss: 2.3542 - val_accuracy: 0.5103
Epoch 6/150
283/283 [==============================] - 29s 102ms/step - loss: 0.6334 - accuracy: 0.8022 - val_loss: 1.2860 - val_accuracy: 0.6233
Epoch 7/150
283/283 [==============================] - 29s 102ms/step - loss: 0.5528 - accuracy: 0.8295 - val_loss: 0.9168 - val_accuracy: 0.7477
Epoch 8/150
283/283 [==============================] - 29s 102ms/step - loss: 0.4464 - accuracy: 0.8623 - val_loss: 1.0062 - val_accuracy: 0.6990
Epoch 9/150
283/283 [==============================] - 29s 102ms/step - loss: 0.3805 - accuracy: 0.8818 - val_loss: 0.7182 - val_accuracy: 0.7990
Epoch 10/150
283/283 [==============================] - 29s 102ms/step - loss: 0.3063 - accuracy: 0.9064 - val_loss: 1.2125 - val_accuracy: 0.7080
Epoch 11/150
283/283 [==============================] - 29s 101ms/step - loss: 0.2997 - accuracy: 0.9119 - val_loss: 0.7861 - val_accuracy: 0.8017
Epoch 12/150
283/283 [==============================] - 29s 102ms/step - loss: 0.2635 - accuracy: 0.9179 - val_loss: 1.2866 - val_accuracy: 0.6813
Epoch 13/150
283/283 [==============================] - 29s 102ms/step - loss: 0.2219 - accuracy: 0.9293 - val_loss: 0.7728 - val_accuracy: 0.7880
Epoch 14/150
283/283 [==============================] - 29s 102ms/step - loss: 0.2383 - accuracy: 0.9297 - val_loss: 1.0445 - val_accuracy: 0.7543
Epoch 15/150
283/283 [==============================] - 29s 102ms/step - loss: 0.2030 - accuracy: 0.9369 - val_loss: 0.3699 - val_accuracy: 0.8977
Epoch 16/150
283/283 [==============================] - 29s 102ms/step - loss: 0.1616 - accuracy: 0.9504 - val_loss: 0.4077 - val_accuracy: 0.8910
Epoch 17/150
283/283 [==============================] - 29s 102ms/step - loss: 0.1639 - accuracy: 0.9518 - val_loss: 1.0256 - val_accuracy: 0.7683
Epoch 18/150
283/283 [==============================] - 29s 101ms/step - loss: 0.1467 - accuracy: 0.9572 - val_loss: 0.3607 - val_accuracy: 0.8957
Epoch 19/150
283/283 [==============================] - 29s 102ms/step - loss: 0.1225 - accuracy: 0.9630 - val_loss: 0.4118 - val_accuracy: 0.8953
Epoch 20/150
283/283 [==============================] - 29s 101ms/step - loss: 0.1781 - accuracy: 0.9451 - val_loss: 0.3018 - val_accuracy: 0.9213
Epoch 21/150
283/283 [==============================] - 29s 101ms/step - loss: 0.1208 - accuracy: 0.9641 - val_loss: 0.4128 - val_accuracy: 0.8870
Epoch 22/150
283/283 [==============================] - 29s 101ms/step - loss: 0.1338 - accuracy: 0.9589 - val_loss: 0.2639 - val_accuracy: 0.9310
Epoch 23/150
283/283 [==============================] - 29s 101ms/step - loss: 0.0822 - accuracy: 0.9751 - val_loss: 0.5764 - val_accuracy: 0.8683
Epoch 24/150
283/283 [==============================] - 29s 101ms/step - loss: 0.1204 - accuracy: 0.9658 - val_loss: 0.3724 - val_accuracy: 0.9070
Epoch 25/150
283/283 [==============================] - 29s 102ms/step - loss: 0.0636 - accuracy: 0.9809 - val_loss: 0.2308 - val_accuracy: 0.9370
Epoch 26/150
283/283 [==============================] - 29s 101ms/step - loss: 0.1271 - accuracy: 0.9615 - val_loss: 0.3110 - val_accuracy: 0.9167
Epoch 27/150
283/283 [==============================] - 29s 102ms/step - loss: 0.1188 - accuracy: 0.9664 - val_loss: 0.4177 - val_accuracy: 0.8963
Epoch 28/150
283/283 [==============================] - 29s 102ms/step - loss: 0.1273 - accuracy: 0.9627 - val_loss: 0.2578 - val_accuracy: 0.9320
Epoch 29/150
283/283 [==============================] - 29s 101ms/step - loss: 0.0670 - accuracy: 0.9809 - val_loss: 0.3640 - val_accuracy: 0.9137
Epoch 30/150
283/283 [==============================] - 29s 101ms/step - loss: 0.0848 - accuracy: 0.9744 - val_loss: 0.9929 - val_accuracy: 0.8233
Epoch 31/150
283/283 [==============================] - 29s 102ms/step - loss: 0.0793 - accuracy: 0.9776 - val_loss: 0.2292 - val_accuracy: 0.9400
Epoch 32/150
283/283 [==============================] - 29s 101ms/step - loss: 0.0840 - accuracy: 0.9767 - val_loss: 0.3639 - val_accuracy: 0.9057
Epoch 33/150
283/283 [==============================] - 30s 104ms/step - loss: 0.0897 - accuracy: 0.9722 - val_loss: 0.2834 - val_accuracy: 0.9293
Epoch 34/150
283/283 [==============================] - 30s 105ms/step - loss: 0.0701 - accuracy: 0.9823 - val_loss: 0.2533 - val_accuracy: 0.9430
Epoch 35/150
283/283 [==============================] - 30s 104ms/step - loss: 0.0965 - accuracy: 0.9721 - val_loss: 0.2141 - val_accuracy: 0.9510
Epoch 36/150
283/283 [==============================] - 30s 105ms/step - loss: 0.0533 - accuracy: 0.9870 - val_loss: 0.2108 - val_accuracy: 0.9503
Epoch 37/150
283/283 [==============================] - 30s 106ms/step - loss: 0.0202 - accuracy: 0.9951 - val_loss: 0.2041 - val_accuracy: 0.9530
Epoch 38/150
283/283 [==============================] - 31s 107ms/step - loss: 0.0228 - accuracy: 0.9951 - val_loss: 0.2113 - val_accuracy: 0.9523
Epoch 39/150
283/283 [==============================] - 34s 119ms/step - loss: 0.0506 - accuracy: 0.9853 - val_loss: 0.3593 - val_accuracy: 0.9183
Epoch 40/150
283/283 [==============================] - 33s 115ms/step - loss: 0.0606 - accuracy: 0.9834 - val_loss: 0.2539 - val_accuracy: 0.9433
Epoch 41/150
283/283 [==============================] - 33s 114ms/step - loss: 0.0306 - accuracy: 0.9919 - val_loss: 0.3426 - val_accuracy: 0.9203
Epoch 42/150
283/283 [==============================] - 43s 149ms/step - loss: 0.0601 - accuracy: 0.9825 - val_loss: 0.3262 - val_accuracy: 0.9270
Epoch 43/150
283/283 [==============================] - 44s 153ms/step - loss: 0.0681 - accuracy: 0.9805 - val_loss: 0.3260 - val_accuracy: 0.9297
Epoch 44/150
283/283 [==============================] - 37s 130ms/step - loss: 0.0266 - accuracy: 0.9931 - val_loss: 0.3613 - val_accuracy: 0.9213
Epoch 45/150
283/283 [==============================] - 33s 117ms/step - loss: 0.0453 - accuracy: 0.9863 - val_loss: 0.2679 - val_accuracy: 0.9383
Epoch 46/150
283/283 [==============================] - 31s 110ms/step - loss: 0.0983 - accuracy: 0.9699 - val_loss: 0.4198 - val_accuracy: 0.9053
Epoch 47/150
283/283 [==============================] - 31s 110ms/step - loss: 0.0299 - accuracy: 0.9919 - val_loss: 0.2019 - val_accuracy: 0.9527
Epoch 48/150
283/283 [==============================] - 31s 107ms/step - loss: 0.0292 - accuracy: 0.9932 - val_loss: 0.2078 - val_accuracy: 0.9540
Epoch 49/150
283/283 [==============================] - 31s 108ms/step - loss: 0.0642 - accuracy: 0.9821 - val_loss: 0.2108 - val_accuracy: 0.9540
Epoch 50/150
283/283 [==============================] - 30s 105ms/step - loss: 0.0287 - accuracy: 0.9924 - val_loss: 0.3471 - val_accuracy: 0.9300
Epoch 51/150
283/283 [==============================] - 30s 105ms/step - loss: 0.0246 - accuracy: 0.9952 - val_loss: 0.2624 - val_accuracy: 0.9383
Epoch 52/150
283/283 [==============================] - 31s 107ms/step - loss: 0.0825 - accuracy: 0.9746 - val_loss: 0.2283 - val_accuracy: 0.9390
Epoch 53/150
283/283 [==============================] - 30s 106ms/step - loss: 0.0358 - accuracy: 0.9911 - val_loss: 0.2587 - val_accuracy: 0.9427
Epoch 54/150
283/283 [==============================] - 31s 107ms/step - loss: 0.0538 - accuracy: 0.9854 - val_loss: 0.2013 - val_accuracy: 0.9507
Epoch 55/150
283/283 [==============================] - 35s 122ms/step - loss: 0.0352 - accuracy: 0.9899 - val_loss: 0.1913 - val_accuracy: 0.9553
Epoch 56/150
283/283 [==============================] - 41s 145ms/step - loss: 0.0237 - accuracy: 0.9940 - val_loss: 0.2169 - val_accuracy: 0.9503
Epoch 57/150
283/283 [==============================] - 37s 131ms/step - loss: 0.0254 - accuracy: 0.9935 - val_loss: 0.3058 - val_accuracy: 0.9293
Epoch 58/150
283/283 [==============================] - 34s 120ms/step - loss: 0.0188 - accuracy: 0.9948 - val_loss: 0.2745 - val_accuracy: 0.9393
Epoch 59/150
283/283 [==============================] - 40s 142ms/step - loss: 0.0427 - accuracy: 0.9885 - val_loss: 0.4339 - val_accuracy: 0.9257
Epoch 60/150
283/283 [==============================] - 36s 128ms/step - loss: 0.0686 - accuracy: 0.9825 - val_loss: 0.3260 - val_accuracy: 0.9320
Epoch 61/150
283/283 [==============================] - 31s 107ms/step - loss: 0.0805 - accuracy: 0.9774 - val_loss: 0.2142 - val_accuracy: 0.9520
Epoch 62/150
283/283 [==============================] - 33s 116ms/step - loss: 0.0179 - accuracy: 0.9962 - val_loss: 0.1659 - val_accuracy: 0.9597
Epoch 63/150
283/283 [==============================] - 38s 135ms/step - loss: 0.0104 - accuracy: 0.9980 - val_loss: 0.1550 - val_accuracy: 0.9617
Epoch 64/150
283/283 [==============================] - 42s 149ms/step - loss: 0.0293 - accuracy: 0.9926 - val_loss: 0.1996 - val_accuracy: 0.9517
Epoch 65/150
283/283 [==============================] - 39s 136ms/step - loss: 0.0181 - accuracy: 0.9962 - val_loss: 0.2459 - val_accuracy: 0.9490
Epoch 66/150
283/283 [==============================] - 31s 109ms/step - loss: 0.0327 - accuracy: 0.9907 - val_loss: 0.3968 - val_accuracy: 0.9103
Epoch 67/150
283/283 [==============================] - 31s 110ms/step - loss: 0.0409 - accuracy: 0.9883 - val_loss: 0.3194 - val_accuracy: 0.9227
Epoch 68/150
283/283 [==============================] - 35s 122ms/step - loss: 0.0138 - accuracy: 0.9975 - val_loss: 0.1712 - val_accuracy: 0.9610
Epoch 69/150
283/283 [==============================] - 35s 122ms/step - loss: 0.0088 - accuracy: 0.9987 - val_loss: 0.1681 - val_accuracy: 0.9647
Epoch 70/150
283/283 [==============================] - 39s 138ms/step - loss: 0.0081 - accuracy: 0.9989 - val_loss: 0.2022 - val_accuracy: 0.9597
Epoch 71/150
283/283 [==============================] - 32s 113ms/step - loss: 0.0369 - accuracy: 0.9901 - val_loss: 0.6708 - val_accuracy: 0.8647
Epoch 72/150
283/283 [==============================] - 31s 108ms/step - loss: 0.0449 - accuracy: 0.9881 - val_loss: 0.4036 - val_accuracy: 0.9180
Epoch 73/150
283/283 [==============================] - 32s 110ms/step - loss: 0.0227 - accuracy: 0.9937 - val_loss: 0.2781 - val_accuracy: 0.9443
Epoch 74/150
283/283 [==============================] - 32s 112ms/step - loss: 0.0227 - accuracy: 0.9941 - val_loss: 0.3525 - val_accuracy: 0.9267
Epoch 75/150
283/283 [==============================] - 32s 113ms/step - loss: 0.0254 - accuracy: 0.9937 - val_loss: 0.2863 - val_accuracy: 0.9390
Epoch 76/150
283/283 [==============================] - 31s 108ms/step - loss: 0.0367 - accuracy: 0.9885 - val_loss: 0.1747 - val_accuracy: 0.9590
Epoch 77/150
283/283 [==============================] - 32s 112ms/step - loss: 0.0267 - accuracy: 0.9930 - val_loss: 0.2391 - val_accuracy: 0.9537
Epoch 78/150
283/283 [==============================] - 31s 107ms/step - loss: 0.0131 - accuracy: 0.9966 - val_loss: 0.1610 - val_accuracy: 0.9597
Epoch 79/150
283/283 [==============================] - 31s 109ms/step - loss: 0.0155 - accuracy: 0.9972 - val_loss: 0.3580 - val_accuracy: 0.9193
Epoch 80/150
283/283 [==============================] - 31s 109ms/step - loss: 0.0206 - accuracy: 0.9951 - val_loss: 0.2150 - val_accuracy: 0.9577
Epoch 81/150
283/283 [==============================] - 32s 113ms/step - loss: 0.0233 - accuracy: 0.9934 - val_loss: 0.4451 - val_accuracy: 0.9150
Epoch 82/150
283/283 [==============================] - 32s 113ms/step - loss: 0.0529 - accuracy: 0.9859 - val_loss: 0.2076 - val_accuracy: 0.9533
Epoch 83/150
283/283 [==============================] - 31s 108ms/step - loss: 0.0597 - accuracy: 0.9832 - val_loss: 0.2463 - val_accuracy: 0.9477
Epoch 84/150
283/283 [==============================] - 31s 107ms/step - loss: 0.0212 - accuracy: 0.9945 - val_loss: 0.1705 - val_accuracy: 0.9630
Epoch 85/150
283/283 [==============================] - 34s 117ms/step - loss: 0.0371 - accuracy: 0.9919 - val_loss: 0.1878 - val_accuracy: 0.9587
Epoch 86/150
283/283 [==============================] - 34s 119ms/step - loss: 0.0369 - accuracy: 0.9895 - val_loss: 0.2304 - val_accuracy: 0.9473
Epoch 87/150
283/283 [==============================] - 33s 114ms/step - loss: 0.0094 - accuracy: 0.9978 - val_loss: 0.1986 - val_accuracy: 0.9580
Epoch 88/150
283/283 [==============================] - 32s 112ms/step - loss: 0.0091 - accuracy: 0.9987 - val_loss: 0.1767 - val_accuracy: 0.9610
Epoch 89/150
283/283 [==============================] - 32s 111ms/step - loss: 0.0196 - accuracy: 0.9949 - val_loss: 0.1793 - val_accuracy: 0.9593
Epoch 90/150
283/283 [==============================] - 34s 118ms/step - loss: 0.0092 - accuracy: 0.9984 - val_loss: 0.1914 - val_accuracy: 0.9623
Epoch 91/150
283/283 [==============================] - 39s 137ms/step - loss: 0.0377 - accuracy: 0.9890 - val_loss: 0.2695 - val_accuracy: 0.9463
Epoch 92/150
283/283 [==============================] - 35s 122ms/step - loss: 0.0511 - accuracy: 0.9859 - val_loss: 0.2105 - val_accuracy: 0.9527
Epoch 93/150
283/283 [==============================] - 35s 124ms/step - loss: 0.0328 - accuracy: 0.9912 - val_loss: 0.1726 - val_accuracy: 0.9653

print(res128_vggmodel.summary())

Model: "sequential_96"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 sequential_97 (Sequential)  (None, 64, 64, 32)        9824      
                                                                 
 sequential_98 (Sequential)  (None, 32, 32, 64)        55936     
                                                                 
 sequential_99 (Sequential)  (None, 16, 16, 128)       370560    
                                                                 
 sequential_100 (Sequential)  (None, 8, 8, 256)        1478400   
                                                                 
 sequential_101 (Sequential)  (None, 4, 4, 256)        1773312   
                                                                 
 dense_60 (Dense)            (None, 4, 4, 4096)        1052672   
                                                                 
 dense_61 (Dense)            (None, 4, 4, 4096)        16781312  
                                                                 
 activity_regularization_21   (None, 4, 4, 4096)       0         
 (ActivityRegularization)                                        
                                                                 
 flatten_21 (Flatten)        (None, 65536)             0         
                                                                 
 dense_62 (Dense)            (None, 15)                983055    
                                                                 
=================================================================
Total params: 22,505,071
Trainable params: 22,500,847
Non-trainable params: 4,224
_________________________________________________________________
None

tf.keras.utils.plot_model(res128_vggmodel, show_shapes=True, dpi=80)

Here, we define a simpler version of the InceptionNet v3 model, using a inception block layout of 3-5-2, We remove one more convolutional layers to simplify it further, but add a regularization layer to stabilize the model. We also implement a dropout layer to prevent combat the model's tendency to overfit.

res128_inceptionmodel = Sequential()
res128_inceptionmodel.add(Conv2D(64, kernel_size = (3,3), strides = 2, padding = "same", activation = "relu"))
res128_inceptionmodel.add(Conv2D(64, kernel_size = (3,3), strides = 1, padding = "same", activation = "relu"))
res128_inceptionmodel.add(MaxPooling2D((2,2)))
res128_inceptionmodel.add(Conv2D(64, kernel_size = (3,3), strides = 1, padding = "same", activation = "relu"))
res128_inceptionmodel.add(Conv2D(64, kernel_size = (3,3), strides = 2, padding = "same", activation = "relu"))
res128_inceptionmodel.add(MaxPooling2D((2,2)))
res128_inceptionmodel.add(inceptionBlock1(64))
res128_inceptionmodel.add(inceptionBlock1(64))
res128_inceptionmodel.add(inceptionBlock1(64))
res128_inceptionmodel.add(inceptionBlock2(num_filters = 64, n = 3))
res128_inceptionmodel.add(inceptionBlock2(num_filters = 64, n = 3))
res128_inceptionmodel.add(inceptionBlock2(num_filters = 64, n = 3))
res128_inceptionmodel.add(inceptionBlock2(num_filters = 64, n = 3))
res128_inceptionmodel.add(inceptionBlock2(num_filters = 64, n = 3))
res128_inceptionmodel.add(inceptionBlock3(num_filters = 64))
res128_inceptionmodel.add(inceptionBlock3(num_filters = 64))
res128_inceptionmodel.add(AveragePooling2D((2,2)))
res128_inceptionmodel.add(Dropout(0.1))
res128_inceptionmodel.add(Dense(2048, activation = "relu"))
res128_inceptionmodel.add(ActivityRegularization(l1 = 0.0005, l2 = 0.0002))
res128_inceptionmodel.add(Flatten())
res128_inceptionmodel.add(Dense(num_classes, activation = "softmax"))
res128_inceptionmodel.compile(optimizer='rmsprop', loss='sparse_categorical_crossentropy', metrics=['accuracy'])

history = res128_inceptionmodel.fit(std_res128_train, validation_data = std_res128_validation, epochs = 150, callbacks = earlyStopper)
cnnf.model_metric_graph(history, res128_inceptionmodel, std_res31_validation, name = "adapted InceptionNet v3", class_names = class_names)

Epoch 1/150
283/283 [==============================] - 35s 88ms/step - loss: 2.6546 - accuracy: 0.1025 - val_loss: 2.7972 - val_accuracy: 0.0667
Epoch 2/150
283/283 [==============================] - 25s 86ms/step - loss: 2.5269 - accuracy: 0.1485 - val_loss: 2.6198 - val_accuracy: 0.1530
Epoch 3/150
283/283 [==============================] - 25s 86ms/step - loss: 2.2093 - accuracy: 0.2829 - val_loss: 2.1044 - val_accuracy: 0.2797
Epoch 4/150
283/283 [==============================] - 24s 85ms/step - loss: 1.8524 - accuracy: 0.4094 - val_loss: 3.3843 - val_accuracy: 0.2267
Epoch 5/150
283/283 [==============================] - 25s 86ms/step - loss: 1.6113 - accuracy: 0.4988 - val_loss: 2.2433 - val_accuracy: 0.3830
Epoch 6/150
283/283 [==============================] - 35s 121ms/step - loss: 1.3612 - accuracy: 0.5773 - val_loss: 2.5690 - val_accuracy: 0.4683
Epoch 7/150
283/283 [==============================] - 34s 120ms/step - loss: 1.1385 - accuracy: 0.6567 - val_loss: 1.3616 - val_accuracy: 0.5973
Epoch 8/150
283/283 [==============================] - 35s 121ms/step - loss: 0.9475 - accuracy: 0.7112 - val_loss: 1.1796 - val_accuracy: 0.6557
Epoch 9/150
283/283 [==============================] - 35s 122ms/step - loss: 0.7960 - accuracy: 0.7656 - val_loss: 3.4560 - val_accuracy: 0.2463
Epoch 10/150
283/283 [==============================] - 35s 120ms/step - loss: 0.6710 - accuracy: 0.8003 - val_loss: 0.8690 - val_accuracy: 0.7510
Epoch 11/150
283/283 [==============================] - 35s 122ms/step - loss: 0.5763 - accuracy: 0.8334 - val_loss: 2.9599 - val_accuracy: 0.4627
Epoch 12/150
283/283 [==============================] - 34s 119ms/step - loss: 0.4907 - accuracy: 0.8654 - val_loss: 0.8416 - val_accuracy: 0.7640
Epoch 13/150
283/283 [==============================] - 34s 119ms/step - loss: 0.4368 - accuracy: 0.8819 - val_loss: 0.5892 - val_accuracy: 0.8433
Epoch 14/150
283/283 [==============================] - 34s 120ms/step - loss: 0.3697 - accuracy: 0.9014 - val_loss: 0.9138 - val_accuracy: 0.7617
Epoch 15/150
283/283 [==============================] - 29s 101ms/step - loss: 0.3228 - accuracy: 0.9169 - val_loss: 1.2821 - val_accuracy: 0.7033
Epoch 16/150
283/283 [==============================] - 27s 93ms/step - loss: 0.3099 - accuracy: 0.9190 - val_loss: 1.0322 - val_accuracy: 0.7300
Epoch 17/150
283/283 [==============================] - 25s 87ms/step - loss: 0.2796 - accuracy: 0.9286 - val_loss: 1.4179 - val_accuracy: 0.7173
Epoch 18/150
283/283 [==============================] - 26s 89ms/step - loss: 0.2518 - accuracy: 0.9358 - val_loss: 0.9803 - val_accuracy: 0.7753
Epoch 19/150
283/283 [==============================] - 26s 90ms/step - loss: 0.2256 - accuracy: 0.9444 - val_loss: 7.0671 - val_accuracy: 0.3923
Epoch 20/150
283/283 [==============================] - 26s 90ms/step - loss: 0.2363 - accuracy: 0.9458 - val_loss: 0.6893 - val_accuracy: 0.8127
Epoch 21/150
283/283 [==============================] - 25s 87ms/step - loss: 0.2179 - accuracy: 0.9486 - val_loss: 0.4290 - val_accuracy: 0.9000
Epoch 22/150
283/283 [==============================] - 25s 86ms/step - loss: 0.1886 - accuracy: 0.9558 - val_loss: 0.8490 - val_accuracy: 0.8257
Epoch 23/150
283/283 [==============================] - 26s 90ms/step - loss: 0.1831 - accuracy: 0.9575 - val_loss: 0.4366 - val_accuracy: 0.8890
Epoch 24/150
283/283 [==============================] - 34s 120ms/step - loss: 0.1667 - accuracy: 0.9615 - val_loss: 0.7386 - val_accuracy: 0.8210
Epoch 25/150
283/283 [==============================] - 34s 119ms/step - loss: 0.1583 - accuracy: 0.9623 - val_loss: 0.4640 - val_accuracy: 0.8840
Epoch 26/150
283/283 [==============================] - 31s 109ms/step - loss: 0.1598 - accuracy: 0.9639 - val_loss: 0.4939 - val_accuracy: 0.8893
Epoch 27/150
283/283 [==============================] - 25s 85ms/step - loss: 0.1633 - accuracy: 0.9643 - val_loss: 0.8082 - val_accuracy: 0.8107
Epoch 28/150
283/283 [==============================] - 25s 85ms/step - loss: 0.1693 - accuracy: 0.9644 - val_loss: 0.6440 - val_accuracy: 0.8370
Epoch 29/150
283/283 [==============================] - 24s 85ms/step - loss: 0.1479 - accuracy: 0.9689 - val_loss: 0.9061 - val_accuracy: 0.8400
Epoch 30/150
283/283 [==============================] - 29s 100ms/step - loss: 0.1551 - accuracy: 0.9688 - val_loss: 0.5892 - val_accuracy: 0.8703
Epoch 31/150
283/283 [==============================] - 33s 114ms/step - loss: 0.1345 - accuracy: 0.9724 - val_loss: 0.5668 - val_accuracy: 0.8623
Epoch 32/150
283/283 [==============================] - 32s 111ms/step - loss: 0.1241 - accuracy: 0.9729 - val_loss: 0.4501 - val_accuracy: 0.8940
Epoch 33/150
283/283 [==============================] - 32s 110ms/step - loss: 0.1265 - accuracy: 0.9726 - val_loss: 0.3528 - val_accuracy: 0.9223
Epoch 34/150
283/283 [==============================] - 32s 111ms/step - loss: 0.1512 - accuracy: 0.9672 - val_loss: 0.4429 - val_accuracy: 0.9007
Epoch 35/150
283/283 [==============================] - 32s 111ms/step - loss: 0.1273 - accuracy: 0.9731 - val_loss: 0.5576 - val_accuracy: 0.8863
Epoch 36/150
283/283 [==============================] - 32s 112ms/step - loss: 0.1331 - accuracy: 0.9711 - val_loss: 0.4371 - val_accuracy: 0.9010
Epoch 37/150
283/283 [==============================] - 34s 117ms/step - loss: 0.1087 - accuracy: 0.9767 - val_loss: 0.4523 - val_accuracy: 0.8937
Epoch 38/150
283/283 [==============================] - 34s 119ms/step - loss: 0.1122 - accuracy: 0.9759 - val_loss: 0.4037 - val_accuracy: 0.9110
Epoch 39/150
283/283 [==============================] - 33s 116ms/step - loss: 0.1128 - accuracy: 0.9746 - val_loss: 0.3977 - val_accuracy: 0.9090
Epoch 40/150
283/283 [==============================] - 33s 116ms/step - loss: 0.1075 - accuracy: 0.9771 - val_loss: 0.4938 - val_accuracy: 0.8913
Epoch 41/150
283/283 [==============================] - 31s 107ms/step - loss: 0.1242 - accuracy: 0.9733 - val_loss: 0.5391 - val_accuracy: 0.8780
Epoch 42/150
283/283 [==============================] - 31s 107ms/step - loss: 0.1223 - accuracy: 0.9715 - val_loss: 0.4469 - val_accuracy: 0.8900
Epoch 43/150
283/283 [==============================] - 30s 106ms/step - loss: 0.1196 - accuracy: 0.9732 - val_loss: 1.1554 - val_accuracy: 0.7793
Epoch 44/150
283/283 [==============================] - 33s 114ms/step - loss: 0.1159 - accuracy: 0.9742 - val_loss: 0.3598 - val_accuracy: 0.9217
Epoch 45/150
283/283 [==============================] - 33s 116ms/step - loss: 0.1096 - accuracy: 0.9762 - val_loss: 0.5892 - val_accuracy: 0.8787
Epoch 46/150
283/283 [==============================] - 32s 110ms/step - loss: 0.1007 - accuracy: 0.9766 - val_loss: 1.3659 - val_accuracy: 0.8467
Epoch 47/150
283/283 [==============================] - 31s 106ms/step - loss: 0.1377 - accuracy: 0.9731 - val_loss: 0.5081 - val_accuracy: 0.8983
Epoch 48/150
283/283 [==============================] - 31s 108ms/step - loss: 0.1194 - accuracy: 0.9741 - val_loss: 0.8553 - val_accuracy: 0.8250
Epoch 49/150
283/283 [==============================] - 31s 109ms/step - loss: 0.1036 - accuracy: 0.9750 - val_loss: 0.4313 - val_accuracy: 0.9077
Epoch 50/150
283/283 [==============================] - 31s 107ms/step - loss: 0.1115 - accuracy: 0.9786 - val_loss: 0.4881 - val_accuracy: 0.8910
Epoch 51/150
283/283 [==============================] - 31s 106ms/step - loss: 0.1206 - accuracy: 0.9716 - val_loss: 0.4809 - val_accuracy: 0.8897
Epoch 52/150
283/283 [==============================] - 32s 112ms/step - loss: 0.0955 - accuracy: 0.9793 - val_loss: 0.6768 - val_accuracy: 0.8787
Epoch 53/150
283/283 [==============================] - 34s 116ms/step - loss: 0.1226 - accuracy: 0.9747 - val_loss: 0.4604 - val_accuracy: 0.8893
Epoch 54/150
283/283 [==============================] - 34s 117ms/step - loss: 0.0932 - accuracy: 0.9784 - val_loss: 0.4904 - val_accuracy: 0.8870
Epoch 55/150
283/283 [==============================] - 31s 108ms/step - loss: 0.1046 - accuracy: 0.9756 - val_loss: 1.4689 - val_accuracy: 0.7683
Epoch 56/150
283/283 [==============================] - 31s 108ms/step - loss: 0.1071 - accuracy: 0.9760 - val_loss: 0.4155 - val_accuracy: 0.9160
Epoch 57/150
283/283 [==============================] - 31s 108ms/step - loss: 0.1096 - accuracy: 0.9761 - val_loss: 0.8019 - val_accuracy: 0.8260
Epoch 58/150
283/283 [==============================] - 32s 112ms/step - loss: 0.1604 - accuracy: 0.9730 - val_loss: 0.5163 - val_accuracy: 0.8887
Epoch 59/150
283/283 [==============================] - 31s 108ms/step - loss: 0.1047 - accuracy: 0.9771 - val_loss: 0.4714 - val_accuracy: 0.9020
Epoch 60/150
283/283 [==============================] - 31s 109ms/step - loss: 0.1074 - accuracy: 0.9753 - val_loss: 0.5976 - val_accuracy: 0.8673
Epoch 61/150
283/283 [==============================] - 32s 111ms/step - loss: 0.1105 - accuracy: 0.9784 - val_loss: 0.5041 - val_accuracy: 0.9053
Epoch 62/150
283/283 [==============================] - 31s 108ms/step - loss: 0.0965 - accuracy: 0.9782 - val_loss: 0.4848 - val_accuracy: 0.8973
Epoch 63/150
283/283 [==============================] - 31s 109ms/step - loss: 0.0995 - accuracy: 0.9782 - val_loss: 0.5448 - val_accuracy: 0.8853

res128_inceptionmodel.summary()

Model: "sequential_72"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 conv2d_162 (Conv2D)         (None, 64, 64, 64)        640       
                                                                 
 conv2d_163 (Conv2D)         (None, 64, 64, 64)        36928     
                                                                 
 max_pooling2d_66 (MaxPoolin  (None, 32, 32, 64)       0         
 g2D)                                                            
                                                                 
 conv2d_164 (Conv2D)         (None, 32, 32, 64)        36928     
                                                                 
 conv2d_165 (Conv2D)         (None, 16, 16, 64)        36928     
                                                                 
 max_pooling2d_67 (MaxPoolin  (None, 8, 8, 64)         0         
 g2D)                                                            
                                                                 
 inception_block1_17 (incept  (None, 8, 8, 256)        127424    
 ionBlock1)                                                      
                                                                 
 inception_block1_18 (incept  (None, 8, 8, 256)        176576    
 ionBlock1)                                                      
                                                                 
 inception_block1_19 (incept  (None, 8, 8, 256)        176576    
 ionBlock1)                                                      
                                                                 
 inception_block2_28 (incept  (None, 8, 8, 256)        139904    
 ionBlock2)                                                      
                                                                 
 inception_block2_29 (incept  (None, 8, 8, 256)        139904    
 ionBlock2)                                                      
                                                                 
 inception_block2_30 (incept  (None, 8, 8, 256)        139904    
 ionBlock2)                                                      
                                                                 
 inception_block2_31 (incept  (None, 8, 8, 256)        139904    
 ionBlock2)                                                      
                                                                 
 inception_block2_32 (incept  (None, 8, 8, 256)        139904    
 ionBlock2)                                                      
                                                                 
 inception_block3_11 (incept  (None, 8, 8, 384)        152128    
 ionBlock3)                                                      
                                                                 
 inception_block3_12 (incept  (None, 8, 8, 384)        184896    
 ionBlock3)                                                      
                                                                 
 average_pooling2d_6 (Averag  (None, 4, 4, 384)        0         
 ePooling2D)                                                     
                                                                 
 dropout_6 (Dropout)         (None, 4, 4, 384)         0         
                                                                 
 dense_44 (Dense)            (None, 4, 4, 2048)        788480    
                                                                 
 activity_regularization_15   (None, 4, 4, 2048)       0         
 (ActivityRegularization)                                        
                                                                 
 flatten_15 (Flatten)        (None, 32768)             0         
                                                                 
 dense_45 (Dense)            (None, 15)                491535    
                                                                 
=================================================================
Total params: 2,908,559
Trainable params: 2,908,559
Non-trainable params: 0
_________________________________________________________________

tf.keras.utils.plot_model(res128_inceptionmodel, show_shapes=True, dpi=80)

finalres128_model_list = [res128_vggmodel, res128_inceptionmodel, baseline128luminosity_model]
finalres128_data_list = [std_res128_validation, std_res128_validation, res128luminosity_validation_data]
name_list = ["Modified VGG-16", "Modified Inception v3", "Baseline CNN"]
cnnf.model_comparison(finalres128_model_list, finalres128_data_list, name_list)

94/94 [==============================] - 5s 46ms/step - loss: 0.1726 - accuracy: 0.9653
94/94 [==============================] - 3s 30ms/step - loss: 0.5448 - accuracy: 0.8853
94/94 [==============================] - 2s 16ms/step - loss: 1.3337 - accuracy: 0.6287

From the above, we can see that the VGG-16 model performs the best, better than both the baseline and the InceptionNet model that we have created.

---

Model Improvement (Explanation)

---

To help us further improve the performance of the model, we shall attempt to fix the class imbalance, and hypertune the parameters of the model to allow it to acheive higher accuracies. Firstly, we shall use augmentation to try to fix the class imbalance. Earlier on, when attempting to create the model, there was a issue where a certain combination of layers and parameters would result in the models predicting only tomatoes (henceforth called Tomato Model). The Tomato Model may have been caused by several factors, one of which being it being the most populous in the dataset. To help the model better predict the less popular classes, we shall augment those classes to give all classes a "fair" chance.

def augment_dataset(dataset, rotation_range=30, zoom_range=0.2, width_shift_range=0.1, height_shift_range=0.1, horizontal_flip=True):
    def augment(image, label):
        # Randomly flip the image horizontally
        if horizontal_flip:
            image = tf.image.random_flip_left_right(image)
        image = tf.image.rot90(image, k=tf.random.uniform(shape=[], minval=0, maxval=4, dtype=tf.int32))
        image = tf.keras.layers.RandomZoom(zoom_range)(image)
        return image, label
    return dataset.map(augment)

aug_res31_train = augment_dataset(std_res31_train)
aug_res128_train = augment_dataset(std_res128_train)

WARNING:tensorflow:Using a while_loop for converting RngReadAndSkip cause there is no registered converter for this op.
WARNING:tensorflow:Using a while_loop for converting Bitcast cause there is no registered converter for this op.
WARNING:tensorflow:Using a while_loop for converting Bitcast cause there is no registered converter for this op.
WARNING:tensorflow:Using a while_loop for converting StatelessRandomUniformV2 cause there is no registered converter for this op.
WARNING:tensorflow:Using a while_loop for converting ImageProjectiveTransformV3 cause there is no registered converter for this op.
WARNING:tensorflow:Using a while_loop for converting RngReadAndSkip cause there is no registered converter for this op.
WARNING:tensorflow:Using a while_loop for converting Bitcast cause there is no registered converter for this op.
WARNING:tensorflow:Using a while_loop for converting Bitcast cause there is no registered converter for this op.
WARNING:tensorflow:Using a while_loop for converting StatelessRandomUniformV2 cause there is no registered converter for this op.
WARNING:tensorflow:Using a while_loop for converting ImageProjectiveTransformV3 cause there is no registered converter for this op.

The figure below shows the effect of augmentation on the (31 x 31) dataset.

plt.figure(figsize=(20, 16))

# Take the first grid_size^2 images from the dataset
for i, (image, _) in enumerate(aug_res31_train.unbatch().take(8 ** 2)):
    # Add a subplot for each image
    plt.subplot(8, 8, i + 1)
    plt.imshow(image.numpy(), cmap = "gray")
    plt.axis("off")
# Display the plot
plt.show()

The figure below shows the effect of augmentation on the (128 x 128) dataset.

plt.figure(figsize=(20, 16))

# Take the first grid_size^2 images from the dataset
for i, (image, _) in enumerate(aug_res128_train.unbatch().take(8 ** 2)):
    # Add a subplot for each image
    plt.subplot(8, 8, i + 1)
    plt.imshow(image.numpy(), cmap = "gray")
    plt.axis("off")
# Display the plot
plt.show()

As we observed earlier, the given dataset is severly imbalanced, with some classes being represented more often than others. To help us combat this, we shall attempt to use class weights to promote a more equal representation of all classes. Class Weights does this by assigning different weights to each class during the training phase, so that it the contribution of each class is balanced. It is typically inversely proportional to the class's frequency in the dataset.

def generate_class_weights(class_series, multi_class=True, one_hot_encoded=False): 
  if multi_class:
    # If class is one hot encoded, transform to categorical labels to use compute_class_weight   
    if one_hot_encoded:
      class_series = np.argmax(class_series, axis=1)
  
    # Compute class weights with sklearn method
    class_labels = np.unique(class_series)
    class_weights = compute_class_weight(class_weight='balanced', classes=class_labels, y=class_series)
    return dict(zip(class_labels, class_weights))
  else:
    # It is neccessary that the multi-label values are one-hot encoded
    mlb = None
    if not one_hot_encoded:
      mlb = MultiLabelBinarizer()
      class_series = mlb.fit_transform(class_series)

    n_samples = len(class_series)
    n_classes = len(class_series[0])

    # Count each class frequency
    class_count = [0] * n_classes
    for classes in class_series:
        for index in range(n_classes):
            if classes[index] != 0:
                class_count[index] += 1
    
    # Compute class weights using balanced method
    class_weights = [n_samples / (n_classes * freq) if freq > 0 else 1 for freq in class_count]
    class_labels = range(len(class_weights)) if mlb is None else mlb.classes_
    return dict(zip(class_labels, class_weights))
  
labels = np.array([label.numpy() for _, label in std_res31_train.unbatch()])
class_weights = generate_class_weights(labels)

print(class_weights)

{0: 0.7716239316239316, 1: 0.835925925925926, 2: 1.3647770219198792, 3: 0.693394777265745, 4: 0.8024888888888889, 5: 1.1965540092776674, 6: 1.7147198480531813, 7: 2.3510416666666667, 8: 1.0253265190232823, 9: 0.7412151067323481, 10: 1.0633686690223794, 11: 1.5964633068081344, 12: 0.7393939393939394, 13: 2.4268817204301074, 14: 0.6302268760907505}

Another improvement that can be done to the model is Random Blockout. Random Blockout works by randomly hiding patches in the image, to force the network to seek for other as many other details in the image for object classification. This can be typically done using black pixels, or by filling it with random values. This is typically done to combat overfitting, and bolster accuracy. However, as our VGG architecture(s) both do not overfit to a large extent, we can skip implementing this improvement.

---

Model Improvement (31 x 31)

---

Now, we shall run the best VGG model layout 3 more times, and with 3 different permutations.

• Class Weights only.
• Augmentation only.
• Both Class Weights and Augmentation applied.

Once those models have been trained, we shall compare it to the base model, with none of these improvments applied, so that we can see how it performs relative to each other.

res31_final = Sequential()
res31_final.add(Input(shape = (31, 31, 1)))
res31_final.add(vgg_block(2, 32))
res31_final.add(vgg_block(2, 64))
res31_final.add(vgg_block(3, 128))
res31_final.add(vgg_block(3, 256))
res31_final.add(Flatten())
res31_final.add(Dense(4096, activation = "relu"))
res31_final.add(Dense(4096, activation = "relu"))
res31_final.add(ActivityRegularization(l1 = 0.0001, l2 = 0.0001))
res31_final.add(Dense(num_classes, activation = "softmax"))
res31_final.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])

res31_final_cw = clone_model(res31_final)
res31_final_aug = clone_model(res31_final)
res31_final_cw_aug = clone_model(res31_final)

res31_final_cw.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])
res31_final_aug.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])
res31_final_cw_aug.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])

history = res31_final_cw.fit(std_res31_train, validation_data = std_res31_validation, epochs = 150, callbacks = earlyStopper, class_weight = class_weights)
cnnf.model_metric_graph(history, res31_final_cw, std_res31_validation, name = "adapted VGG-13 (with class weights applied)", class_names = class_names)

Epoch 1/150
283/283 [==============================] - 13s 39ms/step - loss: 2.4930 - accuracy: 0.1689 - val_loss: 3.6244 - val_accuracy: 0.0730
Epoch 2/150
283/283 [==============================] - 12s 39ms/step - loss: 1.8793 - accuracy: 0.3681 - val_loss: 2.0493 - val_accuracy: 0.3847
Epoch 3/150
283/283 [==============================] - 12s 40ms/step - loss: 1.5249 - accuracy: 0.4973 - val_loss: 1.6812 - val_accuracy: 0.4467
Epoch 4/150
283/283 [==============================] - 12s 39ms/step - loss: 1.1718 - accuracy: 0.6142 - val_loss: 1.3203 - val_accuracy: 0.5823
Epoch 5/150
283/283 [==============================] - 12s 40ms/step - loss: 0.8995 - accuracy: 0.7159 - val_loss: 0.9450 - val_accuracy: 0.7200
Epoch 6/150
283/283 [==============================] - 12s 42ms/step - loss: 0.7108 - accuracy: 0.7754 - val_loss: 1.0075 - val_accuracy: 0.7387
Epoch 7/150
283/283 [==============================] - 12s 41ms/step - loss: 0.6154 - accuracy: 0.8141 - val_loss: 0.9768 - val_accuracy: 0.7180
Epoch 8/150
283/283 [==============================] - 12s 40ms/step - loss: 0.4802 - accuracy: 0.8519 - val_loss: 0.6890 - val_accuracy: 0.8067
Epoch 9/150
283/283 [==============================] - 12s 40ms/step - loss: 0.4280 - accuracy: 0.8667 - val_loss: 0.6552 - val_accuracy: 0.8177
Epoch 10/150
283/283 [==============================] - 12s 41ms/step - loss: 0.3487 - accuracy: 0.8949 - val_loss: 1.0923 - val_accuracy: 0.7283
Epoch 11/150
283/283 [==============================] - 12s 40ms/step - loss: 0.5172 - accuracy: 0.8429 - val_loss: 0.5424 - val_accuracy: 0.8520
Epoch 12/150
283/283 [==============================] - 11s 37ms/step - loss: 0.3209 - accuracy: 0.9045 - val_loss: 0.6538 - val_accuracy: 0.8290
Epoch 13/150
283/283 [==============================] - 11s 38ms/step - loss: 0.2563 - accuracy: 0.9293 - val_loss: 0.4690 - val_accuracy: 0.8797
Epoch 14/150
283/283 [==============================] - 11s 36ms/step - loss: 0.1926 - accuracy: 0.9449 - val_loss: 0.9545 - val_accuracy: 0.7680
Epoch 15/150
283/283 [==============================] - 11s 36ms/step - loss: 0.1834 - accuracy: 0.9482 - val_loss: 0.4439 - val_accuracy: 0.8867
Epoch 16/150
283/283 [==============================] - 11s 37ms/step - loss: 0.1916 - accuracy: 0.9458 - val_loss: 0.4964 - val_accuracy: 0.8703
Epoch 17/150
283/283 [==============================] - 11s 37ms/step - loss: 0.1518 - accuracy: 0.9561 - val_loss: 0.4947 - val_accuracy: 0.8820
Epoch 18/150
283/283 [==============================] - 11s 36ms/step - loss: 0.1755 - accuracy: 0.9530 - val_loss: 0.5209 - val_accuracy: 0.8750
Epoch 19/150
283/283 [==============================] - 11s 37ms/step - loss: 0.1503 - accuracy: 0.9569 - val_loss: 0.4704 - val_accuracy: 0.8787
Epoch 20/150
283/283 [==============================] - 11s 36ms/step - loss: 0.1366 - accuracy: 0.9638 - val_loss: 0.3949 - val_accuracy: 0.9030
Epoch 21/150
283/283 [==============================] - 11s 37ms/step - loss: 0.1223 - accuracy: 0.9640 - val_loss: 0.3906 - val_accuracy: 0.9073
Epoch 22/150
283/283 [==============================] - 11s 36ms/step - loss: 0.1734 - accuracy: 0.9539 - val_loss: 0.4420 - val_accuracy: 0.8870
Epoch 23/150
283/283 [==============================] - 11s 36ms/step - loss: 0.1912 - accuracy: 0.9475 - val_loss: 0.4733 - val_accuracy: 0.8917
Epoch 24/150
283/283 [==============================] - 11s 37ms/step - loss: 0.1322 - accuracy: 0.9653 - val_loss: 0.5020 - val_accuracy: 0.8823
Epoch 25/150
283/283 [==============================] - 11s 38ms/step - loss: 0.1088 - accuracy: 0.9725 - val_loss: 0.4963 - val_accuracy: 0.8823
Epoch 26/150
283/283 [==============================] - 11s 38ms/step - loss: 0.1159 - accuracy: 0.9687 - val_loss: 0.4366 - val_accuracy: 0.9003
Epoch 27/150
283/283 [==============================] - 11s 38ms/step - loss: 0.1040 - accuracy: 0.9720 - val_loss: 0.4794 - val_accuracy: 0.8880
Epoch 28/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0910 - accuracy: 0.9749 - val_loss: 0.4636 - val_accuracy: 0.8997
Epoch 29/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0680 - accuracy: 0.9822 - val_loss: 0.4260 - val_accuracy: 0.9063
Epoch 30/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0624 - accuracy: 0.9845 - val_loss: 0.4731 - val_accuracy: 0.8887
Epoch 31/150
283/283 [==============================] - 11s 37ms/step - loss: 0.1090 - accuracy: 0.9741 - val_loss: 0.6284 - val_accuracy: 0.8500
Epoch 32/150
283/283 [==============================] - 11s 37ms/step - loss: 0.1113 - accuracy: 0.9693 - val_loss: 0.4593 - val_accuracy: 0.8873
Epoch 33/150
283/283 [==============================] - 11s 38ms/step - loss: 0.0871 - accuracy: 0.9788 - val_loss: 0.5398 - val_accuracy: 0.8823
Epoch 34/150
283/283 [==============================] - 11s 37ms/step - loss: 0.1083 - accuracy: 0.9747 - val_loss: 0.3606 - val_accuracy: 0.9263
Epoch 35/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0580 - accuracy: 0.9868 - val_loss: 0.3895 - val_accuracy: 0.9110
Epoch 36/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0657 - accuracy: 0.9838 - val_loss: 0.4114 - val_accuracy: 0.9120
Epoch 37/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0611 - accuracy: 0.9850 - val_loss: 0.8042 - val_accuracy: 0.8223
Epoch 38/150
283/283 [==============================] - 11s 37ms/step - loss: 0.1597 - accuracy: 0.9560 - val_loss: 0.3442 - val_accuracy: 0.9190
Epoch 39/150
283/283 [==============================] - 11s 37ms/step - loss: 0.1158 - accuracy: 0.9711 - val_loss: 0.3928 - val_accuracy: 0.9087
Epoch 40/150
283/283 [==============================] - 11s 38ms/step - loss: 0.0589 - accuracy: 0.9867 - val_loss: 0.4109 - val_accuracy: 0.9110
Epoch 41/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0484 - accuracy: 0.9877 - val_loss: 0.3428 - val_accuracy: 0.9233
Epoch 42/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0409 - accuracy: 0.9897 - val_loss: 0.4386 - val_accuracy: 0.9130
Epoch 43/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0703 - accuracy: 0.9834 - val_loss: 0.3654 - val_accuracy: 0.9253
Epoch 44/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0792 - accuracy: 0.9780 - val_loss: 0.4137 - val_accuracy: 0.9113
Epoch 45/150
283/283 [==============================] - 11s 37ms/step - loss: 0.1004 - accuracy: 0.9756 - val_loss: 0.4368 - val_accuracy: 0.9077
Epoch 46/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0397 - accuracy: 0.9907 - val_loss: 0.4825 - val_accuracy: 0.9013
Epoch 47/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0550 - accuracy: 0.9858 - val_loss: 0.3459 - val_accuracy: 0.9267
Epoch 48/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0442 - accuracy: 0.9895 - val_loss: 0.5641 - val_accuracy: 0.8803
Epoch 49/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0519 - accuracy: 0.9879 - val_loss: 0.4476 - val_accuracy: 0.9167
Epoch 50/150
283/283 [==============================] - 11s 37ms/step - loss: 0.1270 - accuracy: 0.9698 - val_loss: 0.4857 - val_accuracy: 0.8917
Epoch 51/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0417 - accuracy: 0.9895 - val_loss: 0.3637 - val_accuracy: 0.9200
Epoch 52/150
283/283 [==============================] - 12s 39ms/step - loss: 0.0466 - accuracy: 0.9896 - val_loss: 0.4135 - val_accuracy: 0.9120
Epoch 53/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0561 - accuracy: 0.9860 - val_loss: 0.6074 - val_accuracy: 0.8837
Epoch 54/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0567 - accuracy: 0.9863 - val_loss: 0.3741 - val_accuracy: 0.9250
Epoch 55/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0527 - accuracy: 0.9868 - val_loss: 0.4424 - val_accuracy: 0.9073
Epoch 56/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0538 - accuracy: 0.9876 - val_loss: 0.5402 - val_accuracy: 0.8917
Epoch 57/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0357 - accuracy: 0.9932 - val_loss: 0.4244 - val_accuracy: 0.9213
Epoch 58/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0913 - accuracy: 0.9790 - val_loss: 0.4321 - val_accuracy: 0.9100
Epoch 59/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0563 - accuracy: 0.9878 - val_loss: 0.3447 - val_accuracy: 0.9297
Epoch 60/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0443 - accuracy: 0.9896 - val_loss: 0.4158 - val_accuracy: 0.9190
Epoch 61/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0409 - accuracy: 0.9900 - val_loss: 0.4097 - val_accuracy: 0.9227
Epoch 62/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0710 - accuracy: 0.9812 - val_loss: 0.4956 - val_accuracy: 0.8880
Epoch 63/150
283/283 [==============================] - 11s 39ms/step - loss: 0.0668 - accuracy: 0.9853 - val_loss: 0.4945 - val_accuracy: 0.9023
Epoch 64/150
283/283 [==============================] - 11s 38ms/step - loss: 0.0820 - accuracy: 0.9778 - val_loss: 0.4002 - val_accuracy: 0.9190
Epoch 65/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0897 - accuracy: 0.9782 - val_loss: 0.4384 - val_accuracy: 0.9053
Epoch 66/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0585 - accuracy: 0.9868 - val_loss: 0.3849 - val_accuracy: 0.9203
Epoch 67/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0404 - accuracy: 0.9905 - val_loss: 0.3982 - val_accuracy: 0.9163
Epoch 68/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0285 - accuracy: 0.9951 - val_loss: 0.3511 - val_accuracy: 0.9313
Epoch 69/150
283/283 [==============================] - 11s 38ms/step - loss: 0.0351 - accuracy: 0.9912 - val_loss: 0.3626 - val_accuracy: 0.9260
Epoch 70/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0250 - accuracy: 0.9949 - val_loss: 0.3462 - val_accuracy: 0.9363
Epoch 71/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0393 - accuracy: 0.9919 - val_loss: 0.3976 - val_accuracy: 0.9273

history = res31_final_aug.fit(aug_res31_train, validation_data = std_res31_validation, epochs = 150, callbacks = earlyStopper)
cnnf.model_metric_graph(history, res31_final_aug, std_res31_validation, name = "adapted VGG-13 (with augmented images)", class_names = class_names)

Epoch 1/150
283/283 [==============================] - 13s 41ms/step - loss: 2.4398 - accuracy: 0.1981 - val_loss: 3.1321 - val_accuracy: 0.0757
Epoch 2/150
283/283 [==============================] - 12s 39ms/step - loss: 1.9682 - accuracy: 0.3367 - val_loss: 3.3598 - val_accuracy: 0.2123
Epoch 3/150
283/283 [==============================] - 12s 40ms/step - loss: 1.7056 - accuracy: 0.4354 - val_loss: 1.7748 - val_accuracy: 0.3940
Epoch 4/150
283/283 [==============================] - 12s 39ms/step - loss: 1.5320 - accuracy: 0.4940 - val_loss: 1.6474 - val_accuracy: 0.4640
Epoch 5/150
283/283 [==============================] - 12s 40ms/step - loss: 1.4093 - accuracy: 0.5395 - val_loss: 1.7738 - val_accuracy: 0.4403
Epoch 6/150
283/283 [==============================] - 11s 39ms/step - loss: 1.2703 - accuracy: 0.5812 - val_loss: 2.0637 - val_accuracy: 0.4260
Epoch 7/150
283/283 [==============================] - 12s 39ms/step - loss: 1.1731 - accuracy: 0.6100 - val_loss: 1.4396 - val_accuracy: 0.5390
Epoch 8/150
283/283 [==============================] - 12s 39ms/step - loss: 1.0866 - accuracy: 0.6449 - val_loss: 1.3263 - val_accuracy: 0.5770
Epoch 9/150
283/283 [==============================] - 12s 40ms/step - loss: 1.0516 - accuracy: 0.6578 - val_loss: 1.4566 - val_accuracy: 0.5537
Epoch 10/150
283/283 [==============================] - 12s 40ms/step - loss: 0.9537 - accuracy: 0.6906 - val_loss: 1.2470 - val_accuracy: 0.6027
Epoch 11/150
283/283 [==============================] - 12s 39ms/step - loss: 0.8863 - accuracy: 0.7118 - val_loss: 1.0425 - val_accuracy: 0.6533
Epoch 12/150
283/283 [==============================] - 12s 39ms/step - loss: 0.8300 - accuracy: 0.7273 - val_loss: 0.9525 - val_accuracy: 0.6973
Epoch 13/150
283/283 [==============================] - 12s 40ms/step - loss: 0.7830 - accuracy: 0.7492 - val_loss: 0.9494 - val_accuracy: 0.7030
Epoch 14/150
283/283 [==============================] - 12s 39ms/step - loss: 0.7222 - accuracy: 0.7682 - val_loss: 1.1203 - val_accuracy: 0.6617
Epoch 15/150
283/283 [==============================] - 12s 39ms/step - loss: 0.6776 - accuracy: 0.7852 - val_loss: 0.9030 - val_accuracy: 0.7233
Epoch 16/150
283/283 [==============================] - 12s 39ms/step - loss: 0.6463 - accuracy: 0.7947 - val_loss: 0.8872 - val_accuracy: 0.7373
Epoch 17/150
283/283 [==============================] - 11s 39ms/step - loss: 0.6111 - accuracy: 0.8038 - val_loss: 0.9476 - val_accuracy: 0.7190
Epoch 18/150
283/283 [==============================] - 12s 39ms/step - loss: 0.5957 - accuracy: 0.8134 - val_loss: 0.8002 - val_accuracy: 0.7553
Epoch 19/150
283/283 [==============================] - 11s 39ms/step - loss: 0.5362 - accuracy: 0.8343 - val_loss: 0.6834 - val_accuracy: 0.8037
Epoch 20/150
283/283 [==============================] - 12s 40ms/step - loss: 0.5121 - accuracy: 0.8412 - val_loss: 0.9048 - val_accuracy: 0.7380
Epoch 21/150
283/283 [==============================] - 12s 39ms/step - loss: 0.4808 - accuracy: 0.8496 - val_loss: 0.9669 - val_accuracy: 0.7347
Epoch 22/150
283/283 [==============================] - 12s 39ms/step - loss: 0.4541 - accuracy: 0.8604 - val_loss: 0.7536 - val_accuracy: 0.7890
Epoch 23/150
283/283 [==============================] - 12s 39ms/step - loss: 0.4517 - accuracy: 0.8613 - val_loss: 0.8028 - val_accuracy: 0.7520
Epoch 24/150
283/283 [==============================] - 12s 40ms/step - loss: 0.4091 - accuracy: 0.8747 - val_loss: 0.6852 - val_accuracy: 0.7923
Epoch 25/150
283/283 [==============================] - 12s 40ms/step - loss: 0.4034 - accuracy: 0.8744 - val_loss: 0.5779 - val_accuracy: 0.8090
Epoch 26/150
283/283 [==============================] - 12s 40ms/step - loss: 0.3988 - accuracy: 0.8774 - val_loss: 0.8888 - val_accuracy: 0.7587
Epoch 27/150
283/283 [==============================] - 12s 39ms/step - loss: 0.3613 - accuracy: 0.8907 - val_loss: 0.4563 - val_accuracy: 0.8623
Epoch 28/150
283/283 [==============================] - 12s 39ms/step - loss: 0.3674 - accuracy: 0.8864 - val_loss: 0.5585 - val_accuracy: 0.8437
Epoch 29/150
283/283 [==============================] - 12s 40ms/step - loss: 0.3337 - accuracy: 0.8979 - val_loss: 0.7468 - val_accuracy: 0.7810
Epoch 30/150
283/283 [==============================] - 11s 38ms/step - loss: 0.3342 - accuracy: 0.8971 - val_loss: 0.4933 - val_accuracy: 0.8520
Epoch 31/150
283/283 [==============================] - 11s 38ms/step - loss: 0.3179 - accuracy: 0.9018 - val_loss: 0.4569 - val_accuracy: 0.8700
Epoch 32/150
283/283 [==============================] - 11s 39ms/step - loss: 0.2969 - accuracy: 0.9071 - val_loss: 0.4171 - val_accuracy: 0.8753
Epoch 33/150
283/283 [==============================] - 12s 41ms/step - loss: 0.2819 - accuracy: 0.9176 - val_loss: 0.5433 - val_accuracy: 0.8453
Epoch 34/150
283/283 [==============================] - 12s 40ms/step - loss: 0.3211 - accuracy: 0.9002 - val_loss: 0.4284 - val_accuracy: 0.8720
Epoch 35/150
283/283 [==============================] - 12s 40ms/step - loss: 0.2797 - accuracy: 0.9126 - val_loss: 0.5331 - val_accuracy: 0.8547
Epoch 36/150
283/283 [==============================] - 11s 39ms/step - loss: 0.2515 - accuracy: 0.9237 - val_loss: 0.4836 - val_accuracy: 0.8677
Epoch 37/150
283/283 [==============================] - 12s 39ms/step - loss: 0.2969 - accuracy: 0.9094 - val_loss: 0.4221 - val_accuracy: 0.8733
Epoch 38/150
283/283 [==============================] - 12s 39ms/step - loss: 0.2578 - accuracy: 0.9249 - val_loss: 0.3300 - val_accuracy: 0.9057
Epoch 39/150
283/283 [==============================] - 11s 39ms/step - loss: 0.2460 - accuracy: 0.9249 - val_loss: 0.3838 - val_accuracy: 0.8860
Epoch 40/150
283/283 [==============================] - 11s 39ms/step - loss: 0.2280 - accuracy: 0.9290 - val_loss: 0.4468 - val_accuracy: 0.8843
Epoch 41/150
283/283 [==============================] - 11s 39ms/step - loss: 0.2475 - accuracy: 0.9253 - val_loss: 0.3884 - val_accuracy: 0.8853
Epoch 42/150
283/283 [==============================] - 11s 39ms/step - loss: 0.2169 - accuracy: 0.9364 - val_loss: 0.4329 - val_accuracy: 0.8800
Epoch 43/150
283/283 [==============================] - 11s 39ms/step - loss: 0.1938 - accuracy: 0.9420 - val_loss: 0.3714 - val_accuracy: 0.8980
Epoch 44/150
283/283 [==============================] - 12s 39ms/step - loss: 0.2043 - accuracy: 0.9362 - val_loss: 0.4344 - val_accuracy: 0.8780
Epoch 45/150
283/283 [==============================] - 12s 40ms/step - loss: 0.2223 - accuracy: 0.9318 - val_loss: 0.4860 - val_accuracy: 0.8693
Epoch 46/150
283/283 [==============================] - 12s 40ms/step - loss: 0.2395 - accuracy: 0.9303 - val_loss: 0.3626 - val_accuracy: 0.8983
Epoch 47/150
283/283 [==============================] - 12s 40ms/step - loss: 0.2079 - accuracy: 0.9376 - val_loss: 0.5098 - val_accuracy: 0.8593
Epoch 48/150
283/283 [==============================] - 12s 40ms/step - loss: 0.1850 - accuracy: 0.9467 - val_loss: 0.3839 - val_accuracy: 0.8983
Epoch 49/150
283/283 [==============================] - 11s 38ms/step - loss: 0.1656 - accuracy: 0.9500 - val_loss: 0.4774 - val_accuracy: 0.8720
Epoch 50/150
283/283 [==============================] - 11s 38ms/step - loss: 0.1646 - accuracy: 0.9508 - val_loss: 0.4878 - val_accuracy: 0.8713
Epoch 51/150
283/283 [==============================] - 12s 40ms/step - loss: 0.1722 - accuracy: 0.9502 - val_loss: 0.4132 - val_accuracy: 0.8897
Epoch 52/150
283/283 [==============================] - 11s 39ms/step - loss: 0.1826 - accuracy: 0.9487 - val_loss: 0.3501 - val_accuracy: 0.9033
Epoch 53/150
283/283 [==============================] - 12s 39ms/step - loss: 0.1715 - accuracy: 0.9487 - val_loss: 0.5802 - val_accuracy: 0.8707
Epoch 54/150
283/283 [==============================] - 11s 39ms/step - loss: 0.1932 - accuracy: 0.9420 - val_loss: 0.2726 - val_accuracy: 0.9263
Epoch 55/150
283/283 [==============================] - 11s 39ms/step - loss: 0.1720 - accuracy: 0.9482 - val_loss: 0.3100 - val_accuracy: 0.9103
Epoch 56/150
283/283 [==============================] - 11s 39ms/step - loss: 0.1804 - accuracy: 0.9456 - val_loss: 0.3963 - val_accuracy: 0.8970
Epoch 57/150
283/283 [==============================] - 12s 39ms/step - loss: 0.1529 - accuracy: 0.9559 - val_loss: 0.3867 - val_accuracy: 0.8980
Epoch 58/150
283/283 [==============================] - 11s 38ms/step - loss: 0.1599 - accuracy: 0.9544 - val_loss: 0.3995 - val_accuracy: 0.8937
Epoch 59/150
283/283 [==============================] - 11s 39ms/step - loss: 0.1440 - accuracy: 0.9579 - val_loss: 0.2803 - val_accuracy: 0.9257
Epoch 60/150
283/283 [==============================] - 11s 39ms/step - loss: 0.1552 - accuracy: 0.9544 - val_loss: 0.3434 - val_accuracy: 0.9077
Epoch 61/150
283/283 [==============================] - 11s 39ms/step - loss: 0.1676 - accuracy: 0.9521 - val_loss: 0.3224 - val_accuracy: 0.9117
Epoch 62/150
283/283 [==============================] - 11s 39ms/step - loss: 0.1349 - accuracy: 0.9608 - val_loss: 0.3345 - val_accuracy: 0.9140
Epoch 63/150
283/283 [==============================] - 12s 40ms/step - loss: 0.1440 - accuracy: 0.9580 - val_loss: 0.3906 - val_accuracy: 0.8993
Epoch 64/150
283/283 [==============================] - 11s 39ms/step - loss: 0.1292 - accuracy: 0.9605 - val_loss: 0.3295 - val_accuracy: 0.9053
Epoch 65/150
283/283 [==============================] - 11s 39ms/step - loss: 0.1492 - accuracy: 0.9535 - val_loss: 0.3193 - val_accuracy: 0.9177
Epoch 66/150
283/283 [==============================] - 12s 39ms/step - loss: 0.1341 - accuracy: 0.9600 - val_loss: 0.2630 - val_accuracy: 0.9260
Epoch 67/150
283/283 [==============================] - 12s 39ms/step - loss: 0.1542 - accuracy: 0.9568 - val_loss: 0.4331 - val_accuracy: 0.8877
Epoch 68/150
283/283 [==============================] - 11s 39ms/step - loss: 0.1374 - accuracy: 0.9595 - val_loss: 0.2593 - val_accuracy: 0.9257
Epoch 69/150
283/283 [==============================] - 12s 40ms/step - loss: 0.1272 - accuracy: 0.9640 - val_loss: 0.2852 - val_accuracy: 0.9290
Epoch 70/150
283/283 [==============================] - 12s 39ms/step - loss: 0.1205 - accuracy: 0.9631 - val_loss: 0.2339 - val_accuracy: 0.9403
Epoch 71/150
283/283 [==============================] - 11s 38ms/step - loss: 0.1338 - accuracy: 0.9590 - val_loss: 0.3912 - val_accuracy: 0.9030
Epoch 72/150
283/283 [==============================] - 11s 39ms/step - loss: 0.1460 - accuracy: 0.9577 - val_loss: 0.3865 - val_accuracy: 0.8970
Epoch 73/150
283/283 [==============================] - 11s 39ms/step - loss: 0.1195 - accuracy: 0.9671 - val_loss: 0.2593 - val_accuracy: 0.9337
Epoch 74/150
283/283 [==============================] - 11s 38ms/step - loss: 0.1345 - accuracy: 0.9603 - val_loss: 0.3229 - val_accuracy: 0.9217
Epoch 75/150
283/283 [==============================] - 11s 39ms/step - loss: 0.1908 - accuracy: 0.9432 - val_loss: 0.2515 - val_accuracy: 0.9327
Epoch 76/150
283/283 [==============================] - 11s 38ms/step - loss: 0.1100 - accuracy: 0.9675 - val_loss: 0.2455 - val_accuracy: 0.9327
Epoch 77/150
283/283 [==============================] - 11s 39ms/step - loss: 0.1035 - accuracy: 0.9685 - val_loss: 0.2305 - val_accuracy: 0.9380
Epoch 78/150
283/283 [==============================] - 11s 39ms/step - loss: 0.1015 - accuracy: 0.9705 - val_loss: 0.2674 - val_accuracy: 0.9313
Epoch 79/150
283/283 [==============================] - 12s 40ms/step - loss: 0.1087 - accuracy: 0.9695 - val_loss: 0.2712 - val_accuracy: 0.9293
Epoch 80/150
283/283 [==============================] - 12s 42ms/step - loss: 0.1015 - accuracy: 0.9728 - val_loss: 0.3250 - val_accuracy: 0.9230
Epoch 81/150
283/283 [==============================] - 12s 41ms/step - loss: 0.1056 - accuracy: 0.9693 - val_loss: 0.4741 - val_accuracy: 0.8817
Epoch 82/150
283/283 [==============================] - 12s 41ms/step - loss: 0.1211 - accuracy: 0.9665 - val_loss: 0.3901 - val_accuracy: 0.8987
Epoch 83/150
283/283 [==============================] - 12s 41ms/step - loss: 0.1119 - accuracy: 0.9669 - val_loss: 0.2617 - val_accuracy: 0.9340
Epoch 84/150
283/283 [==============================] - 12s 40ms/step - loss: 0.1135 - accuracy: 0.9658 - val_loss: 0.2549 - val_accuracy: 0.9400
Epoch 85/150
283/283 [==============================] - 12s 40ms/step - loss: 0.1044 - accuracy: 0.9698 - val_loss: 0.2705 - val_accuracy: 0.9327
Epoch 86/150
283/283 [==============================] - 12s 40ms/step - loss: 0.1140 - accuracy: 0.9667 - val_loss: 0.3460 - val_accuracy: 0.9150
Epoch 87/150
283/283 [==============================] - 11s 37ms/step - loss: 0.1174 - accuracy: 0.9659 - val_loss: 0.2441 - val_accuracy: 0.9347
Epoch 88/150
283/283 [==============================] - 11s 38ms/step - loss: 0.0848 - accuracy: 0.9763 - val_loss: 0.2476 - val_accuracy: 0.9380
Epoch 89/150
283/283 [==============================] - 11s 38ms/step - loss: 0.0976 - accuracy: 0.9737 - val_loss: 0.3422 - val_accuracy: 0.9143
Epoch 90/150
283/283 [==============================] - 11s 38ms/step - loss: 0.1236 - accuracy: 0.9640 - val_loss: 0.2674 - val_accuracy: 0.9350
Epoch 91/150
283/283 [==============================] - 11s 38ms/step - loss: 0.1048 - accuracy: 0.9731 - val_loss: 0.2862 - val_accuracy: 0.9383
Epoch 92/150
283/283 [==============================] - 11s 38ms/step - loss: 0.1309 - accuracy: 0.9643 - val_loss: 0.2821 - val_accuracy: 0.9257
Epoch 93/150
283/283 [==============================] - 11s 38ms/step - loss: 0.1016 - accuracy: 0.9704 - val_loss: 0.2455 - val_accuracy: 0.9357
Epoch 94/150
283/283 [==============================] - 11s 37ms/step - loss: 0.1114 - accuracy: 0.9677 - val_loss: 0.2208 - val_accuracy: 0.9407
Epoch 95/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0896 - accuracy: 0.9756 - val_loss: 0.3661 - val_accuracy: 0.9177
Epoch 96/150
283/283 [==============================] - 11s 37ms/step - loss: 0.0801 - accuracy: 0.9775 - val_loss: 0.3216 - val_accuracy: 0.9203
Epoch 97/150
283/283 [==============================] - 11s 37ms/step - loss: 0.1135 - accuracy: 0.9683 - val_loss: 0.2149 - val_accuracy: 0.9470
Epoch 98/150
283/283 [==============================] - 11s 38ms/step - loss: 0.0886 - accuracy: 0.9752 - val_loss: 0.2062 - val_accuracy: 0.9490
Epoch 99/150
283/283 [==============================] - 11s 38ms/step - loss: 0.0715 - accuracy: 0.9798 - val_loss: 0.2152 - val_accuracy: 0.9470
Epoch 100/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0810 - accuracy: 0.9766 - val_loss: 0.3809 - val_accuracy: 0.8963
Epoch 101/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0921 - accuracy: 0.9743 - val_loss: 0.2433 - val_accuracy: 0.9353
Epoch 102/150
283/283 [==============================] - 12s 41ms/step - loss: 0.0834 - accuracy: 0.9755 - val_loss: 0.1996 - val_accuracy: 0.9520
Epoch 103/150
283/283 [==============================] - 12s 41ms/step - loss: 0.0838 - accuracy: 0.9763 - val_loss: 0.3722 - val_accuracy: 0.9177
Epoch 104/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0742 - accuracy: 0.9781 - val_loss: 0.2438 - val_accuracy: 0.9383
Epoch 105/150
283/283 [==============================] - 12s 41ms/step - loss: 0.0906 - accuracy: 0.9750 - val_loss: 0.2601 - val_accuracy: 0.9283
Epoch 106/150
283/283 [==============================] - 12s 41ms/step - loss: 0.0721 - accuracy: 0.9787 - val_loss: 0.2436 - val_accuracy: 0.9403
Epoch 107/150
283/283 [==============================] - 12s 41ms/step - loss: 0.0858 - accuracy: 0.9740 - val_loss: 0.2709 - val_accuracy: 0.9353
Epoch 108/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0931 - accuracy: 0.9726 - val_loss: 0.2368 - val_accuracy: 0.9447
Epoch 109/150
283/283 [==============================] - 12s 40ms/step - loss: 0.1092 - accuracy: 0.9703 - val_loss: 0.2127 - val_accuracy: 0.9460
Epoch 110/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0830 - accuracy: 0.9749 - val_loss: 0.2678 - val_accuracy: 0.9287
Epoch 111/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0896 - accuracy: 0.9753 - val_loss: 0.2090 - val_accuracy: 0.9500
Epoch 112/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0629 - accuracy: 0.9824 - val_loss: 0.3180 - val_accuracy: 0.9233
Epoch 113/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0820 - accuracy: 0.9745 - val_loss: 0.2109 - val_accuracy: 0.9483
Epoch 114/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0783 - accuracy: 0.9778 - val_loss: 0.2404 - val_accuracy: 0.9410
Epoch 115/150
283/283 [==============================] - 12s 41ms/step - loss: 0.0849 - accuracy: 0.9781 - val_loss: 0.2048 - val_accuracy: 0.9450
Epoch 116/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0608 - accuracy: 0.9818 - val_loss: 0.2484 - val_accuracy: 0.9380
Epoch 117/150
283/283 [==============================] - 12s 41ms/step - loss: 0.0780 - accuracy: 0.9765 - val_loss: 0.2212 - val_accuracy: 0.9473
Epoch 118/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0696 - accuracy: 0.9793 - val_loss: 0.2367 - val_accuracy: 0.9400
Epoch 119/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0819 - accuracy: 0.9763 - val_loss: 0.3607 - val_accuracy: 0.9107
Epoch 120/150
283/283 [==============================] - 12s 41ms/step - loss: 0.0887 - accuracy: 0.9736 - val_loss: 0.2148 - val_accuracy: 0.9460
Epoch 121/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0885 - accuracy: 0.9760 - val_loss: 0.2497 - val_accuracy: 0.9383
Epoch 122/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0594 - accuracy: 0.9833 - val_loss: 0.2933 - val_accuracy: 0.9273
Epoch 123/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0838 - accuracy: 0.9763 - val_loss: 0.2785 - val_accuracy: 0.9377
Epoch 124/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0892 - accuracy: 0.9719 - val_loss: 0.2149 - val_accuracy: 0.9497
Epoch 125/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0957 - accuracy: 0.9747 - val_loss: 0.1956 - val_accuracy: 0.9493
Epoch 126/150
283/283 [==============================] - 12s 41ms/step - loss: 0.0576 - accuracy: 0.9839 - val_loss: 0.2282 - val_accuracy: 0.9467
Epoch 127/150
283/283 [==============================] - 12s 41ms/step - loss: 0.0678 - accuracy: 0.9813 - val_loss: 0.2319 - val_accuracy: 0.9457
Epoch 128/150
283/283 [==============================] - 12s 41ms/step - loss: 0.0596 - accuracy: 0.9839 - val_loss: 0.2720 - val_accuracy: 0.9413
Epoch 129/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0615 - accuracy: 0.9828 - val_loss: 0.2339 - val_accuracy: 0.9423
Epoch 130/150
283/283 [==============================] - 12s 41ms/step - loss: 0.0876 - accuracy: 0.9744 - val_loss: 0.2406 - val_accuracy: 0.9480
Epoch 131/150
283/283 [==============================] - 12s 41ms/step - loss: 0.1020 - accuracy: 0.9699 - val_loss: 0.2061 - val_accuracy: 0.9530
Epoch 132/150
283/283 [==============================] - 12s 41ms/step - loss: 0.0628 - accuracy: 0.9831 - val_loss: 0.1728 - val_accuracy: 0.9607
Epoch 133/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0525 - accuracy: 0.9854 - val_loss: 0.2360 - val_accuracy: 0.9470
Epoch 134/150
283/283 [==============================] - 12s 41ms/step - loss: 0.0664 - accuracy: 0.9817 - val_loss: 0.1948 - val_accuracy: 0.9607
Epoch 135/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0519 - accuracy: 0.9846 - val_loss: 0.2283 - val_accuracy: 0.9533
Epoch 136/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0568 - accuracy: 0.9839 - val_loss: 0.2471 - val_accuracy: 0.9440
Epoch 137/150
283/283 [==============================] - 12s 41ms/step - loss: 0.0531 - accuracy: 0.9866 - val_loss: 0.2069 - val_accuracy: 0.9550
Epoch 138/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0722 - accuracy: 0.9795 - val_loss: 0.2429 - val_accuracy: 0.9487
Epoch 139/150
283/283 [==============================] - 12s 41ms/step - loss: 0.0651 - accuracy: 0.9817 - val_loss: 0.2891 - val_accuracy: 0.9397
Epoch 140/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0603 - accuracy: 0.9836 - val_loss: 0.2383 - val_accuracy: 0.9377
Epoch 141/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0757 - accuracy: 0.9790 - val_loss: 0.2025 - val_accuracy: 0.9533
Epoch 142/150
283/283 [==============================] - 12s 41ms/step - loss: 0.0648 - accuracy: 0.9813 - val_loss: 0.2236 - val_accuracy: 0.9497
Epoch 143/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0444 - accuracy: 0.9863 - val_loss: 0.2474 - val_accuracy: 0.9487
Epoch 144/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0627 - accuracy: 0.9835 - val_loss: 0.1888 - val_accuracy: 0.9543
Epoch 145/150
283/283 [==============================] - 12s 41ms/step - loss: 0.0532 - accuracy: 0.9849 - val_loss: 0.2598 - val_accuracy: 0.9387
Epoch 146/150
283/283 [==============================] - 12s 41ms/step - loss: 0.0491 - accuracy: 0.9875 - val_loss: 0.2492 - val_accuracy: 0.9420
Epoch 147/150
283/283 [==============================] - 12s 41ms/step - loss: 0.0513 - accuracy: 0.9847 - val_loss: 0.3151 - val_accuracy: 0.9250
Epoch 148/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0715 - accuracy: 0.9809 - val_loss: 0.3110 - val_accuracy: 0.9323
Epoch 149/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0692 - accuracy: 0.9796 - val_loss: 0.2635 - val_accuracy: 0.9383
Epoch 150/150
283/283 [==============================] - 12s 40ms/step - loss: 0.0606 - accuracy: 0.9840 - val_loss: 0.3343 - val_accuracy: 0.9267

history = res31_final_cw_aug.fit(aug_res31_train, validation_data = std_res31_validation, epochs = 150, callbacks = earlyStopper, class_weight = class_weights)
cnnf.model_metric_graph(history, res31_final_cw_aug, std_res31_validation, name = "adapted VGG-13 (with class weights and augmented images applied)", class_names = class_names)

Epoch 1/150
283/283 [==============================] - 13s 40ms/step - loss: 2.5149 - accuracy: 0.1453 - val_loss: 3.0856 - val_accuracy: 0.0737
Epoch 2/150
283/283 [==============================] - 11s 37ms/step - loss: 2.2190 - accuracy: 0.2105 - val_loss: 2.1430 - val_accuracy: 0.2497
Epoch 3/150
283/283 [==============================] - 11s 38ms/step - loss: 2.0479 - accuracy: 0.2962 - val_loss: 2.3048 - val_accuracy: 0.3147
Epoch 4/150
283/283 [==============================] - 11s 37ms/step - loss: 1.8482 - accuracy: 0.3752 - val_loss: 1.9588 - val_accuracy: 0.3890
Epoch 5/150
283/283 [==============================] - 11s 37ms/step - loss: 1.6860 - accuracy: 0.4356 - val_loss: 1.5664 - val_accuracy: 0.5010
Epoch 6/150
283/283 [==============================] - 11s 38ms/step - loss: 1.5316 - accuracy: 0.4940 - val_loss: 1.5872 - val_accuracy: 0.4677
Epoch 7/150
283/283 [==============================] - 11s 37ms/step - loss: 1.4193 - accuracy: 0.5312 - val_loss: 1.9624 - val_accuracy: 0.4530
Epoch 8/150
283/283 [==============================] - 11s 37ms/step - loss: 1.3166 - accuracy: 0.5535 - val_loss: 2.1403 - val_accuracy: 0.3727
Epoch 9/150
283/283 [==============================] - 11s 38ms/step - loss: 1.2070 - accuracy: 0.5986 - val_loss: 1.4844 - val_accuracy: 0.5483
Epoch 10/150
283/283 [==============================] - 11s 37ms/step - loss: 1.1155 - accuracy: 0.6304 - val_loss: 1.6746 - val_accuracy: 0.5100
Epoch 11/150
283/283 [==============================] - 11s 37ms/step - loss: 1.0579 - accuracy: 0.6459 - val_loss: 1.0563 - val_accuracy: 0.6673
Epoch 12/150
283/283 [==============================] - 11s 39ms/step - loss: 1.0047 - accuracy: 0.6643 - val_loss: 0.9559 - val_accuracy: 0.6867
Epoch 13/150
283/283 [==============================] - 12s 39ms/step - loss: 0.9543 - accuracy: 0.6791 - val_loss: 2.0994 - val_accuracy: 0.4927
Epoch 14/150
283/283 [==============================] - 11s 38ms/step - loss: 1.1690 - accuracy: 0.6121 - val_loss: 1.1864 - val_accuracy: 0.6237
Epoch 15/150
283/283 [==============================] - 11s 38ms/step - loss: 0.9137 - accuracy: 0.7004 - val_loss: 1.0238 - val_accuracy: 0.6867
Epoch 16/150
283/283 [==============================] - 11s 38ms/step - loss: 0.8103 - accuracy: 0.7307 - val_loss: 0.7187 - val_accuracy: 0.7800
Epoch 17/150
283/283 [==============================] - 11s 38ms/step - loss: 0.7591 - accuracy: 0.7460 - val_loss: 0.9362 - val_accuracy: 0.7097
Epoch 18/150
283/283 [==============================] - 12s 39ms/step - loss: 0.7571 - accuracy: 0.7482 - val_loss: 0.8355 - val_accuracy: 0.7327
Epoch 19/150
283/283 [==============================] - 11s 38ms/step - loss: 0.6984 - accuracy: 0.7682 - val_loss: 0.8857 - val_accuracy: 0.7363
Epoch 20/150
283/283 [==============================] - 11s 38ms/step - loss: 0.6822 - accuracy: 0.7730 - val_loss: 0.7157 - val_accuracy: 0.7760
Epoch 21/150
283/283 [==============================] - 11s 39ms/step - loss: 0.5831 - accuracy: 0.8105 - val_loss: 0.7471 - val_accuracy: 0.7750
Epoch 22/150
283/283 [==============================] - 11s 38ms/step - loss: 0.5787 - accuracy: 0.8082 - val_loss: 0.5521 - val_accuracy: 0.8283
Epoch 23/150
283/283 [==============================] - 11s 37ms/step - loss: 0.5417 - accuracy: 0.8179 - val_loss: 0.7530 - val_accuracy: 0.7780
Epoch 24/150
283/283 [==============================] - 11s 38ms/step - loss: 0.5181 - accuracy: 0.8270 - val_loss: 0.5584 - val_accuracy: 0.8227
Epoch 25/150
283/283 [==============================] - 11s 38ms/step - loss: 0.5165 - accuracy: 0.8353 - val_loss: 0.5837 - val_accuracy: 0.8213
Epoch 26/150
283/283 [==============================] - 11s 38ms/step - loss: 0.4991 - accuracy: 0.8360 - val_loss: 1.0233 - val_accuracy: 0.7293
Epoch 27/150
283/283 [==============================] - 11s 38ms/step - loss: 0.4335 - accuracy: 0.8581 - val_loss: 0.6213 - val_accuracy: 0.8237
Epoch 28/150
283/283 [==============================] - 11s 38ms/step - loss: 0.4489 - accuracy: 0.8551 - val_loss: 0.7844 - val_accuracy: 0.7833
Epoch 29/150
283/283 [==============================] - 11s 38ms/step - loss: 0.4265 - accuracy: 0.8600 - val_loss: 0.7857 - val_accuracy: 0.7840
Epoch 30/150
283/283 [==============================] - 11s 38ms/step - loss: 0.4494 - accuracy: 0.8530 - val_loss: 0.4980 - val_accuracy: 0.8500
Epoch 31/150
283/283 [==============================] - 11s 38ms/step - loss: 0.3697 - accuracy: 0.8809 - val_loss: 0.3870 - val_accuracy: 0.8857
Epoch 32/150
283/283 [==============================] - 11s 38ms/step - loss: 0.3694 - accuracy: 0.8779 - val_loss: 0.4751 - val_accuracy: 0.8537
Epoch 33/150
283/283 [==============================] - 11s 38ms/step - loss: 0.3596 - accuracy: 0.8806 - val_loss: 0.4119 - val_accuracy: 0.8760
Epoch 34/150
283/283 [==============================] - 11s 38ms/step - loss: 0.3264 - accuracy: 0.8911 - val_loss: 0.9397 - val_accuracy: 0.7527
Epoch 35/150
283/283 [==============================] - 11s 38ms/step - loss: 0.3324 - accuracy: 0.8902 - val_loss: 0.4271 - val_accuracy: 0.8777
Epoch 36/150
283/283 [==============================] - 11s 38ms/step - loss: 0.3408 - accuracy: 0.8947 - val_loss: 0.5157 - val_accuracy: 0.8527
Epoch 37/150
283/283 [==============================] - 11s 38ms/step - loss: 0.3534 - accuracy: 0.8865 - val_loss: 0.4803 - val_accuracy: 0.8707
Epoch 38/150
283/283 [==============================] - 11s 39ms/step - loss: 0.2958 - accuracy: 0.9012 - val_loss: 0.3601 - val_accuracy: 0.8877
Epoch 39/150
283/283 [==============================] - 10s 35ms/step - loss: 0.2881 - accuracy: 0.9107 - val_loss: 0.4203 - val_accuracy: 0.8690
Epoch 40/150
283/283 [==============================] - 10s 35ms/step - loss: 0.2815 - accuracy: 0.9053 - val_loss: 0.5272 - val_accuracy: 0.8590
Epoch 41/150
283/283 [==============================] - 10s 35ms/step - loss: 0.2678 - accuracy: 0.9126 - val_loss: 0.4809 - val_accuracy: 0.8607
Epoch 42/150
283/283 [==============================] - 10s 35ms/step - loss: 0.2727 - accuracy: 0.9109 - val_loss: 0.3893 - val_accuracy: 0.8933
Epoch 43/150
283/283 [==============================] - 10s 35ms/step - loss: 0.2599 - accuracy: 0.9195 - val_loss: 0.5668 - val_accuracy: 0.8490
Epoch 44/150
283/283 [==============================] - 10s 35ms/step - loss: 0.2802 - accuracy: 0.9149 - val_loss: 0.3140 - val_accuracy: 0.9087
Epoch 45/150
283/283 [==============================] - 10s 35ms/step - loss: 0.2512 - accuracy: 0.9180 - val_loss: 0.4343 - val_accuracy: 0.8823
Epoch 46/150
283/283 [==============================] - 10s 35ms/step - loss: 0.2404 - accuracy: 0.9225 - val_loss: 0.3760 - val_accuracy: 0.8950
Epoch 47/150
283/283 [==============================] - 10s 35ms/step - loss: 0.2278 - accuracy: 0.9270 - val_loss: 0.3735 - val_accuracy: 0.8927
Epoch 48/150
283/283 [==============================] - 10s 35ms/step - loss: 0.2096 - accuracy: 0.9329 - val_loss: 0.2525 - val_accuracy: 0.9273
Epoch 49/150
283/283 [==============================] - 10s 35ms/step - loss: 0.2155 - accuracy: 0.9309 - val_loss: 0.3685 - val_accuracy: 0.8893
Epoch 50/150
283/283 [==============================] - 10s 35ms/step - loss: 0.2020 - accuracy: 0.9333 - val_loss: 0.3377 - val_accuracy: 0.9070
Epoch 51/150
283/283 [==============================] - 10s 35ms/step - loss: 0.2058 - accuracy: 0.9354 - val_loss: 0.4763 - val_accuracy: 0.8667
Epoch 52/150
283/283 [==============================] - 10s 35ms/step - loss: 0.2419 - accuracy: 0.9248 - val_loss: 0.3435 - val_accuracy: 0.9000
Epoch 53/150
283/283 [==============================] - 10s 35ms/step - loss: 0.2275 - accuracy: 0.9246 - val_loss: 0.3159 - val_accuracy: 0.9063
Epoch 54/150
283/283 [==============================] - 10s 35ms/step - loss: 0.2216 - accuracy: 0.9304 - val_loss: 0.3660 - val_accuracy: 0.9020
Epoch 55/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1978 - accuracy: 0.9379 - val_loss: 0.3105 - val_accuracy: 0.9133
Epoch 56/150
283/283 [==============================] - 10s 35ms/step - loss: 0.2227 - accuracy: 0.9286 - val_loss: 0.3057 - val_accuracy: 0.9107
Epoch 57/150
283/283 [==============================] - 10s 35ms/step - loss: 0.2303 - accuracy: 0.9258 - val_loss: 0.3883 - val_accuracy: 0.8900
Epoch 58/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1750 - accuracy: 0.9421 - val_loss: 0.2655 - val_accuracy: 0.9253
Epoch 59/150
283/283 [==============================] - 10s 35ms/step - loss: 0.2428 - accuracy: 0.9220 - val_loss: 0.3117 - val_accuracy: 0.9067
Epoch 60/150
283/283 [==============================] - 11s 36ms/step - loss: 0.1583 - accuracy: 0.9499 - val_loss: 0.2866 - val_accuracy: 0.9243
Epoch 61/150
283/283 [==============================] - 10s 35ms/step - loss: 0.2419 - accuracy: 0.9262 - val_loss: 0.2558 - val_accuracy: 0.9300
Epoch 62/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1864 - accuracy: 0.9380 - val_loss: 0.2828 - val_accuracy: 0.9223
Epoch 63/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1524 - accuracy: 0.9495 - val_loss: 0.3207 - val_accuracy: 0.9163
Epoch 64/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1556 - accuracy: 0.9533 - val_loss: 0.2774 - val_accuracy: 0.9250
Epoch 65/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1495 - accuracy: 0.9548 - val_loss: 0.2605 - val_accuracy: 0.9373
Epoch 66/150
283/283 [==============================] - 11s 36ms/step - loss: 0.1595 - accuracy: 0.9466 - val_loss: 0.2504 - val_accuracy: 0.9330
Epoch 67/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1534 - accuracy: 0.9528 - val_loss: 0.3194 - val_accuracy: 0.9110
Epoch 68/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1536 - accuracy: 0.9529 - val_loss: 0.3813 - val_accuracy: 0.8997
Epoch 69/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1926 - accuracy: 0.9433 - val_loss: 0.2857 - val_accuracy: 0.9207
Epoch 70/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1357 - accuracy: 0.9527 - val_loss: 0.2566 - val_accuracy: 0.9300
Epoch 71/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1391 - accuracy: 0.9566 - val_loss: 0.2806 - val_accuracy: 0.9273
Epoch 72/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1489 - accuracy: 0.9515 - val_loss: 0.5356 - val_accuracy: 0.8610
Epoch 73/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1474 - accuracy: 0.9525 - val_loss: 0.2196 - val_accuracy: 0.9447
Epoch 74/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1370 - accuracy: 0.9537 - val_loss: 0.4135 - val_accuracy: 0.8980
Epoch 75/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1428 - accuracy: 0.9550 - val_loss: 0.2680 - val_accuracy: 0.9270
Epoch 76/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1522 - accuracy: 0.9546 - val_loss: 0.2470 - val_accuracy: 0.9367
Epoch 77/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1219 - accuracy: 0.9595 - val_loss: 0.2811 - val_accuracy: 0.9260
Epoch 78/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1312 - accuracy: 0.9598 - val_loss: 0.2948 - val_accuracy: 0.9243
Epoch 79/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1429 - accuracy: 0.9560 - val_loss: 0.2648 - val_accuracy: 0.9290
Epoch 80/150
283/283 [==============================] - 11s 37ms/step - loss: 0.1308 - accuracy: 0.9599 - val_loss: 0.2565 - val_accuracy: 0.9300
Epoch 81/150
283/283 [==============================] - 11s 36ms/step - loss: 0.1240 - accuracy: 0.9610 - val_loss: 0.4957 - val_accuracy: 0.8673
Epoch 82/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1225 - accuracy: 0.9592 - val_loss: 0.2080 - val_accuracy: 0.9457
Epoch 83/150
283/283 [==============================] - 11s 37ms/step - loss: 0.1868 - accuracy: 0.9446 - val_loss: 0.2859 - val_accuracy: 0.9283
Epoch 84/150
283/283 [==============================] - 11s 36ms/step - loss: 0.1284 - accuracy: 0.9598 - val_loss: 0.2112 - val_accuracy: 0.9447
Epoch 85/150
283/283 [==============================] - 11s 36ms/step - loss: 0.1457 - accuracy: 0.9575 - val_loss: 0.2161 - val_accuracy: 0.9437
Epoch 86/150
283/283 [==============================] - 11s 36ms/step - loss: 0.1260 - accuracy: 0.9608 - val_loss: 0.3359 - val_accuracy: 0.9110
Epoch 87/150
283/283 [==============================] - 11s 36ms/step - loss: 0.1255 - accuracy: 0.9649 - val_loss: 0.2728 - val_accuracy: 0.9260
Epoch 88/150
283/283 [==============================] - 11s 36ms/step - loss: 0.0968 - accuracy: 0.9701 - val_loss: 0.4084 - val_accuracy: 0.8993
Epoch 89/150
283/283 [==============================] - 11s 36ms/step - loss: 0.1103 - accuracy: 0.9639 - val_loss: 0.1984 - val_accuracy: 0.9483
Epoch 90/150
283/283 [==============================] - 11s 36ms/step - loss: 0.1077 - accuracy: 0.9664 - val_loss: 0.2646 - val_accuracy: 0.9363
Epoch 91/150
283/283 [==============================] - 11s 36ms/step - loss: 0.1179 - accuracy: 0.9665 - val_loss: 0.2522 - val_accuracy: 0.9333
Epoch 92/150
283/283 [==============================] - 11s 36ms/step - loss: 0.1243 - accuracy: 0.9634 - val_loss: 0.3316 - val_accuracy: 0.9163
Epoch 93/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1188 - accuracy: 0.9617 - val_loss: 0.2566 - val_accuracy: 0.9323
Epoch 94/150
283/283 [==============================] - 10s 35ms/step - loss: 0.0981 - accuracy: 0.9677 - val_loss: 0.4244 - val_accuracy: 0.9033
Epoch 95/150
283/283 [==============================] - 10s 35ms/step - loss: 0.0938 - accuracy: 0.9711 - val_loss: 0.2364 - val_accuracy: 0.9433
Epoch 96/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1153 - accuracy: 0.9634 - val_loss: 0.2433 - val_accuracy: 0.9283
Epoch 97/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1042 - accuracy: 0.9658 - val_loss: 0.4172 - val_accuracy: 0.8870
Epoch 98/150
283/283 [==============================] - 10s 35ms/step - loss: 0.0904 - accuracy: 0.9729 - val_loss: 0.3618 - val_accuracy: 0.9133
Epoch 99/150
283/283 [==============================] - 11s 36ms/step - loss: 0.0957 - accuracy: 0.9722 - val_loss: 0.2508 - val_accuracy: 0.9330
Epoch 100/150
283/283 [==============================] - 11s 35ms/step - loss: 0.0949 - accuracy: 0.9695 - val_loss: 0.3296 - val_accuracy: 0.9220
Epoch 101/150
283/283 [==============================] - 11s 36ms/step - loss: 0.1316 - accuracy: 0.9601 - val_loss: 0.3467 - val_accuracy: 0.9093
Epoch 102/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1065 - accuracy: 0.9677 - val_loss: 0.2188 - val_accuracy: 0.9450
Epoch 103/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1047 - accuracy: 0.9652 - val_loss: 0.2321 - val_accuracy: 0.9410
Epoch 104/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1094 - accuracy: 0.9662 - val_loss: 0.3980 - val_accuracy: 0.8990
Epoch 105/150
283/283 [==============================] - 10s 35ms/step - loss: 0.0991 - accuracy: 0.9726 - val_loss: 0.2598 - val_accuracy: 0.9350
Epoch 106/150
283/283 [==============================] - 10s 35ms/step - loss: 0.0665 - accuracy: 0.9795 - val_loss: 0.2365 - val_accuracy: 0.9420
Epoch 107/150
283/283 [==============================] - 10s 35ms/step - loss: 0.0843 - accuracy: 0.9744 - val_loss: 0.2373 - val_accuracy: 0.9370
Epoch 108/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1062 - accuracy: 0.9685 - val_loss: 0.3143 - val_accuracy: 0.9163
Epoch 109/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1069 - accuracy: 0.9682 - val_loss: 0.2349 - val_accuracy: 0.9427
Epoch 110/150
283/283 [==============================] - 10s 35ms/step - loss: 0.0953 - accuracy: 0.9734 - val_loss: 0.3617 - val_accuracy: 0.9153
Epoch 111/150
283/283 [==============================] - 10s 35ms/step - loss: 0.0714 - accuracy: 0.9782 - val_loss: 0.3045 - val_accuracy: 0.9193
Epoch 112/150
283/283 [==============================] - 10s 35ms/step - loss: 0.0765 - accuracy: 0.9783 - val_loss: 0.2720 - val_accuracy: 0.9353
Epoch 113/150
283/283 [==============================] - 11s 36ms/step - loss: 0.0848 - accuracy: 0.9742 - val_loss: 0.2748 - val_accuracy: 0.9277
Epoch 114/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1098 - accuracy: 0.9657 - val_loss: 0.2828 - val_accuracy: 0.9257
Epoch 115/150
283/283 [==============================] - 10s 35ms/step - loss: 0.0797 - accuracy: 0.9766 - val_loss: 0.3115 - val_accuracy: 0.9253
Epoch 116/150
283/283 [==============================] - 10s 35ms/step - loss: 0.1159 - accuracy: 0.9608 - val_loss: 0.2272 - val_accuracy: 0.9400
Epoch 117/150
283/283 [==============================] - 10s 35ms/step - loss: 0.0870 - accuracy: 0.9706 - val_loss: 0.2055 - val_accuracy: 0.9533
Epoch 118/150
283/283 [==============================] - 10s 35ms/step - loss: 0.0770 - accuracy: 0.9764 - val_loss: 0.2620 - val_accuracy: 0.9497
Epoch 119/150
283/283 [==============================] - 11s 36ms/step - loss: 0.0984 - accuracy: 0.9732 - val_loss: 0.2034 - val_accuracy: 0.9473

improvement31_model_list = [res31_vggmodel, res31_final_cw, res31_final_aug, res31_final_cw_aug]
improvement31_data_list = [std_res31_validation, std_res31_validation, std_res31_validation, std_res31_validation]
name_list = ["VGG Model with no improvements", "Class Weights Applied", "Augmentation Applied", "Class Weights and Augmentation Applied"]
cnnf.model_comparison(improvement31_model_list, improvement31_data_list, name_list)

94/94 [==============================] - 2s 16ms/step - loss: 0.4841 - accuracy: 0.8953
94/94 [==============================] - 2s 15ms/step - loss: 0.3976 - accuracy: 0.9273
94/94 [==============================] - 2s 15ms/step - loss: 0.3343 - accuracy: 0.9267
94/94 [==============================] - 2s 15ms/step - loss: 0.2034 - accuracy: 0.9473

From the above, we can see that all of the improvements that we attempted have had a improvement on the model's performance, with the application of Class Weights and Augmentation having the most profound improvement, with accuracy and loss both being the best performing among all the 4 permutations. Hence, we shall now select this model and hypertune it.

From this, we can conclude that augmentation is useful for the 31 x 31 image input size.

We shall save the model that we want to hypertune first, to allow the hypertuning to be seperate from the other models, saving on computational resources.

res31_final_cw_aug.save('res31_best.h5')

---

Model Improvement (128 x 128)

---

Now, we shall run the best VGG model layout 3 more times, and with 3 different permutations.

• Class Weights only.
• Augmentation only.
• Both Class Weights and Augmentation applied.

Once those models have been trained, we shall compare it to the base model, with none of these improvments applied, so that we can see how it performs relative to each other.

res128_final = Sequential()
res128_final.add(Input(shape = (128, 128, 1)))
res128_final.add(vgg_block(2, 32))
res128_final.add(vgg_block(2, 64))
res128_final.add(vgg_block(3, 128))
res128_final.add(vgg_block(3, 256))
res128_final.add(vgg_block(3, 256))
res128_final.add(Dense(4096, activation = "relu"))
res128_final.add(Dense(4096, activation = "relu"))
res128_final.add(ActivityRegularization(l1 = 0.00001, l2 = 0.00001))
res128_final.add(Flatten())
res128_final.add(Dense(num_classes, activation = "softmax"))
res128_final.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])

res128_final_cw = clone_model(res128_final)
res128_final_aug = clone_model(res128_final)
res128_final_cw_aug = clone_model(res128_final)

res128_final_cw.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])
res128_final_aug.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])
res128_final_cw_aug.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])

history = res128_final_cw.fit(std_res128_train, validation_data = std_res128_validation, epochs = 150, callbacks = earlyStopper, class_weight = class_weights)
cnnf.model_metric_graph(history, res128_final_cw, std_res31_validation, name = "adapted VGG-16 (with class weights applied)", class_names = class_names)

Epoch 1/150
283/283 [==============================] - 53s 179ms/step - loss: 2.3629 - accuracy: 0.2233 - val_loss: 3.3180 - val_accuracy: 0.0667
Epoch 2/150
283/283 [==============================] - 50s 177ms/step - loss: 1.6899 - accuracy: 0.4466 - val_loss: 2.0270 - val_accuracy: 0.4310
Epoch 3/150
283/283 [==============================] - 51s 179ms/step - loss: 1.3144 - accuracy: 0.5714 - val_loss: 1.5473 - val_accuracy: 0.5450
Epoch 4/150
283/283 [==============================] - 50s 176ms/step - loss: 1.0891 - accuracy: 0.6382 - val_loss: 1.5125 - val_accuracy: 0.5380
Epoch 5/150
283/283 [==============================] - 50s 177ms/step - loss: 0.9119 - accuracy: 0.6942 - val_loss: 1.7319 - val_accuracy: 0.4853
Epoch 6/150
283/283 [==============================] - 50s 176ms/step - loss: 0.7636 - accuracy: 0.7484 - val_loss: 1.0949 - val_accuracy: 0.6930
Epoch 7/150
283/283 [==============================] - 50s 176ms/step - loss: 0.6525 - accuracy: 0.7871 - val_loss: 1.1454 - val_accuracy: 0.6877
Epoch 8/150
283/283 [==============================] - 50s 176ms/step - loss: 0.5617 - accuracy: 0.8108 - val_loss: 0.8808 - val_accuracy: 0.7130
Epoch 9/150
283/283 [==============================] - 50s 177ms/step - loss: 0.4703 - accuracy: 0.8443 - val_loss: 0.6634 - val_accuracy: 0.7907
Epoch 10/150
283/283 [==============================] - 52s 181ms/step - loss: 0.4420 - accuracy: 0.8539 - val_loss: 0.6856 - val_accuracy: 0.8153
Epoch 11/150
283/283 [==============================] - 51s 178ms/step - loss: 0.3386 - accuracy: 0.8848 - val_loss: 0.7040 - val_accuracy: 0.8057
Epoch 12/150
283/283 [==============================] - 50s 177ms/step - loss: 0.2894 - accuracy: 0.9031 - val_loss: 0.4977 - val_accuracy: 0.8597
Epoch 13/150
283/283 [==============================] - 50s 176ms/step - loss: 0.3191 - accuracy: 0.8929 - val_loss: 0.4525 - val_accuracy: 0.8717
Epoch 14/150
283/283 [==============================] - 51s 179ms/step - loss: 0.2634 - accuracy: 0.9129 - val_loss: 0.5365 - val_accuracy: 0.8400
Epoch 15/150
283/283 [==============================] - 51s 179ms/step - loss: 0.3445 - accuracy: 0.8890 - val_loss: 0.4246 - val_accuracy: 0.8687
Epoch 16/150
283/283 [==============================] - 51s 178ms/step - loss: 0.2129 - accuracy: 0.9335 - val_loss: 0.3143 - val_accuracy: 0.9133
Epoch 17/150
283/283 [==============================] - 51s 180ms/step - loss: 0.1800 - accuracy: 0.9405 - val_loss: 0.5567 - val_accuracy: 0.8617
Epoch 18/150
283/283 [==============================] - 51s 179ms/step - loss: 0.2519 - accuracy: 0.9198 - val_loss: 0.4602 - val_accuracy: 0.8780
Epoch 19/150
283/283 [==============================] - 51s 179ms/step - loss: 0.1488 - accuracy: 0.9504 - val_loss: 0.4120 - val_accuracy: 0.8960
Epoch 20/150
283/283 [==============================] - 51s 177ms/step - loss: 0.1131 - accuracy: 0.9630 - val_loss: 0.3844 - val_accuracy: 0.8977
Epoch 21/150
283/283 [==============================] - 50s 177ms/step - loss: 0.0933 - accuracy: 0.9673 - val_loss: 0.5925 - val_accuracy: 0.8610
Epoch 22/150
283/283 [==============================] - 51s 178ms/step - loss: 0.1206 - accuracy: 0.9599 - val_loss: 0.3952 - val_accuracy: 0.9010
Epoch 23/150
283/283 [==============================] - 52s 181ms/step - loss: 0.2825 - accuracy: 0.9114 - val_loss: 0.5023 - val_accuracy: 0.8720
Epoch 24/150
283/283 [==============================] - 52s 182ms/step - loss: 0.2050 - accuracy: 0.9311 - val_loss: 0.3410 - val_accuracy: 0.9100
Epoch 25/150
283/283 [==============================] - 53s 187ms/step - loss: 0.0992 - accuracy: 0.9659 - val_loss: 0.3153 - val_accuracy: 0.9187
Epoch 26/150
283/283 [==============================] - 51s 180ms/step - loss: 0.0931 - accuracy: 0.9700 - val_loss: 0.3916 - val_accuracy: 0.9060
Epoch 27/150
283/283 [==============================] - 51s 179ms/step - loss: 0.1013 - accuracy: 0.9681 - val_loss: 0.2619 - val_accuracy: 0.9353
Epoch 28/150
283/283 [==============================] - 51s 179ms/step - loss: 0.0530 - accuracy: 0.9839 - val_loss: 0.2774 - val_accuracy: 0.9397
Epoch 29/150
283/283 [==============================] - 51s 179ms/step - loss: 0.0755 - accuracy: 0.9774 - val_loss: 0.6310 - val_accuracy: 0.8490
Epoch 30/150
283/283 [==============================] - 51s 180ms/step - loss: 0.0728 - accuracy: 0.9770 - val_loss: 0.5339 - val_accuracy: 0.8737
Epoch 31/150
283/283 [==============================] - 52s 181ms/step - loss: 0.0952 - accuracy: 0.9715 - val_loss: 0.7391 - val_accuracy: 0.8370
Epoch 32/150
283/283 [==============================] - 51s 179ms/step - loss: 0.0972 - accuracy: 0.9678 - val_loss: 0.3420 - val_accuracy: 0.9237
Epoch 33/150
283/283 [==============================] - 51s 178ms/step - loss: 0.0901 - accuracy: 0.9729 - val_loss: 0.6628 - val_accuracy: 0.8283
Epoch 34/150
283/283 [==============================] - 51s 179ms/step - loss: 0.2124 - accuracy: 0.9329 - val_loss: 0.2955 - val_accuracy: 0.9243
Epoch 35/150
283/283 [==============================] - 51s 178ms/step - loss: 0.0657 - accuracy: 0.9795 - val_loss: 0.2813 - val_accuracy: 0.9303
Epoch 36/150
283/283 [==============================] - 53s 186ms/step - loss: 0.0378 - accuracy: 0.9879 - val_loss: 0.2979 - val_accuracy: 0.9367
Epoch 37/150
283/283 [==============================] - 54s 189ms/step - loss: 0.0785 - accuracy: 0.9747 - val_loss: 0.3274 - val_accuracy: 0.9300
Epoch 38/150
283/283 [==============================] - 54s 189ms/step - loss: 0.1107 - accuracy: 0.9648 - val_loss: 0.3199 - val_accuracy: 0.9267
Epoch 39/150
283/283 [==============================] - 53s 186ms/step - loss: 0.0665 - accuracy: 0.9782 - val_loss: 0.2493 - val_accuracy: 0.9453
Epoch 40/150
283/283 [==============================] - 53s 184ms/step - loss: 0.0833 - accuracy: 0.9746 - val_loss: 0.2715 - val_accuracy: 0.9353
Epoch 41/150
283/283 [==============================] - 53s 187ms/step - loss: 0.0517 - accuracy: 0.9850 - val_loss: 0.3346 - val_accuracy: 0.9250
Epoch 42/150
283/283 [==============================] - 53s 186ms/step - loss: 0.0445 - accuracy: 0.9868 - val_loss: 0.2910 - val_accuracy: 0.9343
Epoch 43/150
283/283 [==============================] - 57s 201ms/step - loss: 0.0293 - accuracy: 0.9924 - val_loss: 0.2413 - val_accuracy: 0.9423
Epoch 44/150
283/283 [==============================] - 56s 197ms/step - loss: 0.0522 - accuracy: 0.9849 - val_loss: 0.2749 - val_accuracy: 0.9417
Epoch 45/150
283/283 [==============================] - 56s 197ms/step - loss: 0.0806 - accuracy: 0.9771 - val_loss: 0.2593 - val_accuracy: 0.9403
Epoch 46/150
283/283 [==============================] - 54s 188ms/step - loss: 0.0274 - accuracy: 0.9928 - val_loss: 0.3371 - val_accuracy: 0.9313
Epoch 47/150
283/283 [==============================] - 51s 180ms/step - loss: 0.1179 - accuracy: 0.9638 - val_loss: 0.3566 - val_accuracy: 0.9160
Epoch 48/150
283/283 [==============================] - 52s 181ms/step - loss: 0.0557 - accuracy: 0.9824 - val_loss: 0.2881 - val_accuracy: 0.9343
Epoch 49/150
283/283 [==============================] - 52s 184ms/step - loss: 0.0412 - accuracy: 0.9880 - val_loss: 0.4669 - val_accuracy: 0.8910
Epoch 50/150
283/283 [==============================] - 52s 183ms/step - loss: 0.0850 - accuracy: 0.9764 - val_loss: 0.2608 - val_accuracy: 0.9467
Epoch 51/150
283/283 [==============================] - 52s 182ms/step - loss: 0.0333 - accuracy: 0.9917 - val_loss: 0.2857 - val_accuracy: 0.9403
Epoch 52/150
283/283 [==============================] - 51s 180ms/step - loss: 0.0757 - accuracy: 0.9764 - val_loss: 0.3082 - val_accuracy: 0.9343
Epoch 53/150
283/283 [==============================] - 52s 183ms/step - loss: 0.0345 - accuracy: 0.9895 - val_loss: 0.3128 - val_accuracy: 0.9263
Epoch 54/150
283/283 [==============================] - 52s 181ms/step - loss: 0.1106 - accuracy: 0.9640 - val_loss: 0.2644 - val_accuracy: 0.9327
Epoch 55/150
283/283 [==============================] - 52s 182ms/step - loss: 0.0569 - accuracy: 0.9811 - val_loss: 0.2570 - val_accuracy: 0.9437
Epoch 56/150
283/283 [==============================] - 52s 183ms/step - loss: 0.0401 - accuracy: 0.9869 - val_loss: 0.2285 - val_accuracy: 0.9510
Epoch 57/150
283/283 [==============================] - 51s 180ms/step - loss: 0.0279 - accuracy: 0.9925 - val_loss: 0.2883 - val_accuracy: 0.9360
Epoch 58/150
283/283 [==============================] - 52s 181ms/step - loss: 0.0413 - accuracy: 0.9875 - val_loss: 0.3526 - val_accuracy: 0.9207
Epoch 59/150
283/283 [==============================] - 52s 181ms/step - loss: 0.0329 - accuracy: 0.9893 - val_loss: 0.2770 - val_accuracy: 0.9453
Epoch 60/150
283/283 [==============================] - 51s 180ms/step - loss: 0.0359 - accuracy: 0.9890 - val_loss: 0.2392 - val_accuracy: 0.9500
Epoch 61/150
283/283 [==============================] - 52s 181ms/step - loss: 0.0287 - accuracy: 0.9922 - val_loss: 0.3243 - val_accuracy: 0.9373
Epoch 62/150
283/283 [==============================] - 52s 182ms/step - loss: 0.0878 - accuracy: 0.9749 - val_loss: 0.2909 - val_accuracy: 0.9343
Epoch 63/150
283/283 [==============================] - 51s 180ms/step - loss: 0.0237 - accuracy: 0.9937 - val_loss: 0.2827 - val_accuracy: 0.9367
Epoch 64/150
283/283 [==============================] - 52s 183ms/step - loss: 0.0254 - accuracy: 0.9931 - val_loss: 0.4546 - val_accuracy: 0.9000
Epoch 65/150
283/283 [==============================] - 51s 180ms/step - loss: 0.0225 - accuracy: 0.9938 - val_loss: 0.1997 - val_accuracy: 0.9570
Epoch 66/150
283/283 [==============================] - 52s 182ms/step - loss: 0.0136 - accuracy: 0.9970 - val_loss: 0.2603 - val_accuracy: 0.9443
Epoch 67/150
283/283 [==============================] - 52s 182ms/step - loss: 0.0801 - accuracy: 0.9733 - val_loss: 0.2942 - val_accuracy: 0.9387
Epoch 68/150
283/283 [==============================] - 53s 186ms/step - loss: 0.0368 - accuracy: 0.9899 - val_loss: 0.2751 - val_accuracy: 0.9407
Epoch 69/150
283/283 [==============================] - 55s 191ms/step - loss: 0.0376 - accuracy: 0.9890 - val_loss: 0.2301 - val_accuracy: 0.9540
Epoch 70/150
283/283 [==============================] - 54s 189ms/step - loss: 0.0406 - accuracy: 0.9900 - val_loss: 0.2795 - val_accuracy: 0.9417
Epoch 71/150
283/283 [==============================] - 54s 190ms/step - loss: 0.0819 - accuracy: 0.9760 - val_loss: 0.2494 - val_accuracy: 0.9503
Epoch 72/150
283/283 [==============================] - 54s 191ms/step - loss: 0.0272 - accuracy: 0.9920 - val_loss: 0.2267 - val_accuracy: 0.9597
Epoch 73/150
283/283 [==============================] - 52s 183ms/step - loss: 0.0185 - accuracy: 0.9950 - val_loss: 0.2734 - val_accuracy: 0.9473
Epoch 74/150
283/283 [==============================] - 52s 183ms/step - loss: 0.0193 - accuracy: 0.9944 - val_loss: 0.2074 - val_accuracy: 0.9590
Epoch 75/150
283/283 [==============================] - 52s 181ms/step - loss: 0.0135 - accuracy: 0.9969 - val_loss: 0.2891 - val_accuracy: 0.9347
Epoch 76/150
283/283 [==============================] - 52s 181ms/step - loss: 0.0550 - accuracy: 0.9836 - val_loss: 0.2542 - val_accuracy: 0.9453
Epoch 77/150
283/283 [==============================] - 51s 180ms/step - loss: 0.0477 - accuracy: 0.9869 - val_loss: 0.2344 - val_accuracy: 0.9503
Epoch 78/150
283/283 [==============================] - 52s 181ms/step - loss: 0.0327 - accuracy: 0.9908 - val_loss: 0.1674 - val_accuracy: 0.9627
Epoch 79/150
283/283 [==============================] - 52s 181ms/step - loss: 0.0196 - accuracy: 0.9955 - val_loss: 0.3024 - val_accuracy: 0.9400
Epoch 80/150
283/283 [==============================] - 51s 180ms/step - loss: 0.0290 - accuracy: 0.9907 - val_loss: 0.2208 - val_accuracy: 0.9523
Epoch 81/150
283/283 [==============================] - 52s 181ms/step - loss: 0.0404 - accuracy: 0.9878 - val_loss: 0.2688 - val_accuracy: 0.9437
Epoch 82/150
283/283 [==============================] - 51s 180ms/step - loss: 0.0235 - accuracy: 0.9937 - val_loss: 0.3683 - val_accuracy: 0.9283
Epoch 83/150
283/283 [==============================] - 51s 179ms/step - loss: 0.0115 - accuracy: 0.9975 - val_loss: 0.2035 - val_accuracy: 0.9587
Epoch 84/150
283/283 [==============================] - 52s 181ms/step - loss: 0.0322 - accuracy: 0.9885 - val_loss: 0.2437 - val_accuracy: 0.9513
Epoch 85/150
283/283 [==============================] - 52s 184ms/step - loss: 0.0402 - accuracy: 0.9894 - val_loss: 0.2252 - val_accuracy: 0.9573
Epoch 86/150
283/283 [==============================] - 52s 181ms/step - loss: 0.0201 - accuracy: 0.9939 - val_loss: 0.2198 - val_accuracy: 0.9537
Epoch 87/150
283/283 [==============================] - 52s 181ms/step - loss: 0.0130 - accuracy: 0.9965 - val_loss: 0.1982 - val_accuracy: 0.9583
Epoch 88/150
283/283 [==============================] - 52s 182ms/step - loss: 0.0147 - accuracy: 0.9957 - val_loss: 0.2318 - val_accuracy: 0.9530
Epoch 89/150
283/283 [==============================] - 51s 180ms/step - loss: 0.0199 - accuracy: 0.9945 - val_loss: 0.2571 - val_accuracy: 0.9527
Epoch 90/150
283/283 [==============================] - 52s 181ms/step - loss: 0.0192 - accuracy: 0.9949 - val_loss: 0.3401 - val_accuracy: 0.9293
Epoch 91/150
283/283 [==============================] - 52s 182ms/step - loss: 0.0615 - accuracy: 0.9822 - val_loss: 0.4606 - val_accuracy: 0.8917
Epoch 92/150
283/283 [==============================] - 52s 183ms/step - loss: 0.0204 - accuracy: 0.9944 - val_loss: 0.2258 - val_accuracy: 0.9557
Epoch 93/150
283/283 [==============================] - 52s 183ms/step - loss: 0.0252 - accuracy: 0.9921 - val_loss: 0.2199 - val_accuracy: 0.9563
Epoch 94/150
283/283 [==============================] - 52s 182ms/step - loss: 0.0094 - accuracy: 0.9979 - val_loss: 0.2388 - val_accuracy: 0.9503
Epoch 95/150
283/283 [==============================] - 52s 182ms/step - loss: 0.0818 - accuracy: 0.9801 - val_loss: 0.2465 - val_accuracy: 0.9477
Epoch 96/150
283/283 [==============================] - 52s 182ms/step - loss: 0.0154 - accuracy: 0.9963 - val_loss: 0.2886 - val_accuracy: 0.9450
Epoch 97/150
283/283 [==============================] - 52s 182ms/step - loss: 0.0262 - accuracy: 0.9926 - val_loss: 0.2160 - val_accuracy: 0.9563
Epoch 98/150
283/283 [==============================] - 52s 181ms/step - loss: 0.0507 - accuracy: 0.9864 - val_loss: 0.2800 - val_accuracy: 0.9437
Epoch 99/150
283/283 [==============================] - 52s 181ms/step - loss: 0.0488 - accuracy: 0.9840 - val_loss: 0.2641 - val_accuracy: 0.9440
Epoch 100/150
283/283 [==============================] - 52s 182ms/step - loss: 0.0169 - accuracy: 0.9960 - val_loss: 0.2034 - val_accuracy: 0.9600
Epoch 101/150
283/283 [==============================] - 52s 181ms/step - loss: 0.0090 - accuracy: 0.9986 - val_loss: 0.1896 - val_accuracy: 0.9607
Epoch 102/150
283/283 [==============================] - 52s 181ms/step - loss: 0.0369 - accuracy: 0.9895 - val_loss: 0.2093 - val_accuracy: 0.9570
Epoch 103/150
283/283 [==============================] - 52s 182ms/step - loss: 0.0190 - accuracy: 0.9946 - val_loss: 0.2460 - val_accuracy: 0.9493
Epoch 104/150
283/283 [==============================] - 52s 182ms/step - loss: 0.0199 - accuracy: 0.9947 - val_loss: 0.2078 - val_accuracy: 0.9623
Epoch 105/150
283/283 [==============================] - 52s 183ms/step - loss: 0.0094 - accuracy: 0.9978 - val_loss: 0.2042 - val_accuracy: 0.9643
Epoch 106/150
283/283 [==============================] - 52s 181ms/step - loss: 0.0307 - accuracy: 0.9910 - val_loss: 0.2227 - val_accuracy: 0.9507
Epoch 107/150
283/283 [==============================] - 55s 192ms/step - loss: 0.0312 - accuracy: 0.9894 - val_loss: 0.2051 - val_accuracy: 0.9603
Epoch 108/150
283/283 [==============================] - 52s 181ms/step - loss: 0.0101 - accuracy: 0.9978 - val_loss: 0.2479 - val_accuracy: 0.9527

history = res128_final_aug.fit(aug_res128_train, validation_data = std_res128_validation, epochs = 150, callbacks = earlyStopper)
cnnf.model_metric_graph(history, res128_final_aug, std_res31_validation, name = "adapted VGG-16 (with augmented images)", class_names = class_names)

Epoch 1/150
283/283 [==============================] - 58s 198ms/step - loss: 2.4222 - accuracy: 0.2112 - val_loss: 3.4832 - val_accuracy: 0.0667
Epoch 2/150
283/283 [==============================] - 56s 197ms/step - loss: 1.8777 - accuracy: 0.3677 - val_loss: 2.4124 - val_accuracy: 0.2070
Epoch 3/150
283/283 [==============================] - 56s 196ms/step - loss: 1.6010 - accuracy: 0.4747 - val_loss: 2.6549 - val_accuracy: 0.2847
Epoch 4/150
283/283 [==============================] - 56s 197ms/step - loss: 1.4333 - accuracy: 0.5239 - val_loss: 2.4287 - val_accuracy: 0.3647
Epoch 5/150
283/283 [==============================] - 56s 197ms/step - loss: 1.3196 - accuracy: 0.5670 - val_loss: 1.3495 - val_accuracy: 0.5473
Epoch 6/150
283/283 [==============================] - 56s 197ms/step - loss: 1.1703 - accuracy: 0.6141 - val_loss: 1.7751 - val_accuracy: 0.4633
Epoch 7/150
283/283 [==============================] - 56s 197ms/step - loss: 1.1085 - accuracy: 0.6400 - val_loss: 1.8938 - val_accuracy: 0.4240
Epoch 8/150
283/283 [==============================] - 56s 198ms/step - loss: 0.9717 - accuracy: 0.6843 - val_loss: 2.3660 - val_accuracy: 0.4623
Epoch 9/150
283/283 [==============================] - 56s 196ms/step - loss: 0.9012 - accuracy: 0.7141 - val_loss: 1.2971 - val_accuracy: 0.6227
Epoch 10/150
283/283 [==============================] - 56s 197ms/step - loss: 0.8414 - accuracy: 0.7292 - val_loss: 2.4885 - val_accuracy: 0.4290
Epoch 11/150
283/283 [==============================] - 56s 198ms/step - loss: 0.7244 - accuracy: 0.7683 - val_loss: 1.5485 - val_accuracy: 0.5453
Epoch 12/150
283/283 [==============================] - 56s 198ms/step - loss: 0.6947 - accuracy: 0.7780 - val_loss: 1.0486 - val_accuracy: 0.6657
Epoch 13/150
283/283 [==============================] - 56s 197ms/step - loss: 0.6312 - accuracy: 0.8023 - val_loss: 0.8162 - val_accuracy: 0.7563
Epoch 14/150
283/283 [==============================] - 56s 198ms/step - loss: 0.5585 - accuracy: 0.8208 - val_loss: 0.7568 - val_accuracy: 0.7670
Epoch 15/150
283/283 [==============================] - 56s 197ms/step - loss: 0.5660 - accuracy: 0.8160 - val_loss: 0.7518 - val_accuracy: 0.7610
Epoch 16/150
283/283 [==============================] - 56s 197ms/step - loss: 0.5067 - accuracy: 0.8406 - val_loss: 1.0937 - val_accuracy: 0.6810
Epoch 17/150
283/283 [==============================] - 59s 207ms/step - loss: 0.4534 - accuracy: 0.8500 - val_loss: 1.2870 - val_accuracy: 0.6883
Epoch 18/150
283/283 [==============================] - 57s 198ms/step - loss: 0.4493 - accuracy: 0.8531 - val_loss: 0.5108 - val_accuracy: 0.8390
Epoch 19/150
283/283 [==============================] - 56s 197ms/step - loss: 0.3931 - accuracy: 0.8763 - val_loss: 0.6002 - val_accuracy: 0.8253
Epoch 20/150
283/283 [==============================] - 56s 197ms/step - loss: 0.3460 - accuracy: 0.8889 - val_loss: 0.9029 - val_accuracy: 0.7457
Epoch 21/150
283/283 [==============================] - 56s 197ms/step - loss: 0.3702 - accuracy: 0.8868 - val_loss: 0.4910 - val_accuracy: 0.8530
Epoch 22/150
283/283 [==============================] - 56s 198ms/step - loss: 0.3264 - accuracy: 0.9012 - val_loss: 0.6350 - val_accuracy: 0.8187
Epoch 23/150
283/283 [==============================] - 56s 197ms/step - loss: 0.3166 - accuracy: 0.8994 - val_loss: 0.4755 - val_accuracy: 0.8570
Epoch 24/150
283/283 [==============================] - 56s 196ms/step - loss: 0.3047 - accuracy: 0.9077 - val_loss: 0.4231 - val_accuracy: 0.8700
Epoch 25/150
283/283 [==============================] - 56s 195ms/step - loss: 0.2992 - accuracy: 0.9084 - val_loss: 0.2325 - val_accuracy: 0.9263
Epoch 26/150
283/283 [==============================] - 56s 195ms/step - loss: 0.2672 - accuracy: 0.9184 - val_loss: 0.9854 - val_accuracy: 0.7580
Epoch 27/150
283/283 [==============================] - 56s 198ms/step - loss: 0.2520 - accuracy: 0.9255 - val_loss: 0.6002 - val_accuracy: 0.8443
Epoch 28/150
283/283 [==============================] - 57s 198ms/step - loss: 0.2324 - accuracy: 0.9284 - val_loss: 0.3202 - val_accuracy: 0.8963
Epoch 29/150
283/283 [==============================] - 56s 197ms/step - loss: 0.2265 - accuracy: 0.9294 - val_loss: 0.6791 - val_accuracy: 0.8303
Epoch 30/150
283/283 [==============================] - 57s 199ms/step - loss: 0.2239 - accuracy: 0.9313 - val_loss: 0.2150 - val_accuracy: 0.9390
Epoch 31/150
283/283 [==============================] - 57s 199ms/step - loss: 0.1979 - accuracy: 0.9356 - val_loss: 0.4534 - val_accuracy: 0.8563
Epoch 32/150
283/283 [==============================] - 57s 198ms/step - loss: 0.1981 - accuracy: 0.9386 - val_loss: 1.0555 - val_accuracy: 0.7063
Epoch 33/150
283/283 [==============================] - 56s 198ms/step - loss: 0.1903 - accuracy: 0.9403 - val_loss: 1.0971 - val_accuracy: 0.7640
Epoch 34/150
283/283 [==============================] - 56s 196ms/step - loss: 0.2054 - accuracy: 0.9385 - val_loss: 0.5920 - val_accuracy: 0.8407
Epoch 35/150
283/283 [==============================] - 56s 197ms/step - loss: 0.1677 - accuracy: 0.9473 - val_loss: 0.7191 - val_accuracy: 0.8080
Epoch 36/150
283/283 [==============================] - 56s 196ms/step - loss: 0.1876 - accuracy: 0.9438 - val_loss: 0.5993 - val_accuracy: 0.8380
Epoch 37/150
283/283 [==============================] - 56s 197ms/step - loss: 0.1551 - accuracy: 0.9521 - val_loss: 0.2037 - val_accuracy: 0.9420
Epoch 38/150
283/283 [==============================] - 56s 198ms/step - loss: 0.1623 - accuracy: 0.9495 - val_loss: 0.3490 - val_accuracy: 0.9077
Epoch 39/150
283/283 [==============================] - 57s 198ms/step - loss: 0.1430 - accuracy: 0.9572 - val_loss: 0.2632 - val_accuracy: 0.9197
Epoch 40/150
283/283 [==============================] - 56s 197ms/step - loss: 0.1412 - accuracy: 0.9557 - val_loss: 0.3002 - val_accuracy: 0.9130
Epoch 41/150
283/283 [==============================] - 56s 197ms/step - loss: 0.1273 - accuracy: 0.9593 - val_loss: 0.5748 - val_accuracy: 0.8693
Epoch 42/150
283/283 [==============================] - 56s 198ms/step - loss: 0.1463 - accuracy: 0.9545 - val_loss: 0.3179 - val_accuracy: 0.9157
Epoch 43/150
283/283 [==============================] - 56s 197ms/step - loss: 0.1389 - accuracy: 0.9543 - val_loss: 0.4218 - val_accuracy: 0.8803
Epoch 44/150
283/283 [==============================] - 56s 197ms/step - loss: 0.1447 - accuracy: 0.9569 - val_loss: 0.1932 - val_accuracy: 0.9493
Epoch 45/150
283/283 [==============================] - 56s 196ms/step - loss: 0.1323 - accuracy: 0.9575 - val_loss: 0.2755 - val_accuracy: 0.9163
Epoch 46/150
283/283 [==============================] - 56s 196ms/step - loss: 0.1267 - accuracy: 0.9608 - val_loss: 0.2934 - val_accuracy: 0.9267
Epoch 47/150
283/283 [==============================] - 56s 196ms/step - loss: 0.1251 - accuracy: 0.9606 - val_loss: 0.2220 - val_accuracy: 0.9467
Epoch 48/150
283/283 [==============================] - 56s 196ms/step - loss: 0.1126 - accuracy: 0.9656 - val_loss: 0.1647 - val_accuracy: 0.9563
Epoch 49/150
283/283 [==============================] - 56s 197ms/step - loss: 0.1418 - accuracy: 0.9553 - val_loss: 0.1860 - val_accuracy: 0.9560
Epoch 50/150
283/283 [==============================] - 56s 197ms/step - loss: 0.1275 - accuracy: 0.9613 - val_loss: 0.5491 - val_accuracy: 0.8623
Epoch 51/150
283/283 [==============================] - 56s 197ms/step - loss: 0.1110 - accuracy: 0.9649 - val_loss: 0.1730 - val_accuracy: 0.9563
Epoch 52/150
283/283 [==============================] - 56s 197ms/step - loss: 0.1188 - accuracy: 0.9620 - val_loss: 0.2401 - val_accuracy: 0.9347
Epoch 53/150
283/283 [==============================] - 56s 198ms/step - loss: 0.0970 - accuracy: 0.9695 - val_loss: 0.3167 - val_accuracy: 0.9210
Epoch 54/150
283/283 [==============================] - 58s 202ms/step - loss: 0.0994 - accuracy: 0.9694 - val_loss: 0.2470 - val_accuracy: 0.9313
Epoch 55/150
283/283 [==============================] - 56s 196ms/step - loss: 0.1151 - accuracy: 0.9663 - val_loss: 0.1210 - val_accuracy: 0.9643
Epoch 56/150
283/283 [==============================] - 56s 196ms/step - loss: 0.1096 - accuracy: 0.9675 - val_loss: 0.1423 - val_accuracy: 0.9593
Epoch 57/150
283/283 [==============================] - 55s 194ms/step - loss: 0.0804 - accuracy: 0.9757 - val_loss: 0.1355 - val_accuracy: 0.9623
Epoch 58/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0864 - accuracy: 0.9754 - val_loss: 0.1665 - val_accuracy: 0.9563
Epoch 59/150
283/283 [==============================] - 55s 195ms/step - loss: 0.1069 - accuracy: 0.9691 - val_loss: 0.1947 - val_accuracy: 0.9507
Epoch 60/150
283/283 [==============================] - 55s 194ms/step - loss: 0.0906 - accuracy: 0.9722 - val_loss: 0.1477 - val_accuracy: 0.9577
Epoch 61/150
283/283 [==============================] - 55s 194ms/step - loss: 0.0885 - accuracy: 0.9737 - val_loss: 0.1305 - val_accuracy: 0.9663
Epoch 62/150
283/283 [==============================] - 56s 195ms/step - loss: 0.1052 - accuracy: 0.9681 - val_loss: 0.0827 - val_accuracy: 0.9730
Epoch 63/150
283/283 [==============================] - 56s 196ms/step - loss: 0.0765 - accuracy: 0.9750 - val_loss: 0.1062 - val_accuracy: 0.9680
Epoch 64/150
283/283 [==============================] - 55s 194ms/step - loss: 0.0900 - accuracy: 0.9721 - val_loss: 0.1793 - val_accuracy: 0.9530
Epoch 65/150
283/283 [==============================] - 55s 195ms/step - loss: 0.0788 - accuracy: 0.9749 - val_loss: 0.0865 - val_accuracy: 0.9810
Epoch 66/150
283/283 [==============================] - 55s 194ms/step - loss: 0.1044 - accuracy: 0.9731 - val_loss: 0.2020 - val_accuracy: 0.9450
Epoch 67/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0992 - accuracy: 0.9688 - val_loss: 0.1543 - val_accuracy: 0.9570
Epoch 68/150
283/283 [==============================] - 55s 194ms/step - loss: 0.0733 - accuracy: 0.9776 - val_loss: 0.1040 - val_accuracy: 0.9713
Epoch 69/150
283/283 [==============================] - 55s 194ms/step - loss: 0.0603 - accuracy: 0.9801 - val_loss: 0.1224 - val_accuracy: 0.9667
Epoch 70/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0677 - accuracy: 0.9802 - val_loss: 0.1683 - val_accuracy: 0.9547
Epoch 71/150
283/283 [==============================] - 55s 195ms/step - loss: 0.0582 - accuracy: 0.9831 - val_loss: 0.2791 - val_accuracy: 0.9267
Epoch 72/150
283/283 [==============================] - 58s 205ms/step - loss: 0.1021 - accuracy: 0.9701 - val_loss: 0.1065 - val_accuracy: 0.9750
Epoch 73/150
283/283 [==============================] - 58s 197ms/step - loss: 0.0459 - accuracy: 0.9862 - val_loss: 0.1241 - val_accuracy: 0.9693
Epoch 74/150
283/283 [==============================] - 55s 193ms/step - loss: 0.0687 - accuracy: 0.9787 - val_loss: 0.1492 - val_accuracy: 0.9550
Epoch 75/150
283/283 [==============================] - 56s 196ms/step - loss: 0.0755 - accuracy: 0.9771 - val_loss: 0.2038 - val_accuracy: 0.9453
Epoch 76/150
283/283 [==============================] - 55s 194ms/step - loss: 0.0794 - accuracy: 0.9772 - val_loss: 0.1972 - val_accuracy: 0.9450
Epoch 77/150
283/283 [==============================] - 55s 194ms/step - loss: 0.0603 - accuracy: 0.9822 - val_loss: 0.1846 - val_accuracy: 0.9500
Epoch 78/150
283/283 [==============================] - 55s 194ms/step - loss: 0.0840 - accuracy: 0.9742 - val_loss: 0.1316 - val_accuracy: 0.9640
Epoch 79/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0560 - accuracy: 0.9844 - val_loss: 0.0781 - val_accuracy: 0.9760
Epoch 80/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0965 - accuracy: 0.9702 - val_loss: 0.1219 - val_accuracy: 0.9660
Epoch 81/150
283/283 [==============================] - 56s 196ms/step - loss: 0.0535 - accuracy: 0.9839 - val_loss: 0.0975 - val_accuracy: 0.9713
Epoch 82/150
283/283 [==============================] - 55s 194ms/step - loss: 0.0619 - accuracy: 0.9804 - val_loss: 0.1325 - val_accuracy: 0.9640
Epoch 83/150
283/283 [==============================] - 56s 196ms/step - loss: 0.0538 - accuracy: 0.9842 - val_loss: 0.0987 - val_accuracy: 0.9757
Epoch 84/150
283/283 [==============================] - 56s 196ms/step - loss: 0.0593 - accuracy: 0.9805 - val_loss: 0.1192 - val_accuracy: 0.9660
Epoch 85/150
283/283 [==============================] - 56s 196ms/step - loss: 0.0703 - accuracy: 0.9785 - val_loss: 0.0905 - val_accuracy: 0.9757
Epoch 86/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0456 - accuracy: 0.9865 - val_loss: 0.1438 - val_accuracy: 0.9647
Epoch 87/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0556 - accuracy: 0.9817 - val_loss: 0.0963 - val_accuracy: 0.9737
Epoch 88/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0631 - accuracy: 0.9812 - val_loss: 0.0567 - val_accuracy: 0.9850
Epoch 89/150
283/283 [==============================] - 55s 194ms/step - loss: 0.0458 - accuracy: 0.9864 - val_loss: 0.1353 - val_accuracy: 0.9663
Epoch 90/150
283/283 [==============================] - 55s 194ms/step - loss: 0.0750 - accuracy: 0.9774 - val_loss: 0.1365 - val_accuracy: 0.9643
Epoch 91/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0503 - accuracy: 0.9854 - val_loss: 0.1098 - val_accuracy: 0.9673
Epoch 92/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0485 - accuracy: 0.9853 - val_loss: 0.1387 - val_accuracy: 0.9643
Epoch 93/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0530 - accuracy: 0.9837 - val_loss: 0.1271 - val_accuracy: 0.9673
Epoch 94/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0578 - accuracy: 0.9824 - val_loss: 0.0674 - val_accuracy: 0.9813
Epoch 95/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0499 - accuracy: 0.9850 - val_loss: 0.1024 - val_accuracy: 0.9700
Epoch 96/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0637 - accuracy: 0.9798 - val_loss: 0.1325 - val_accuracy: 0.9650
Epoch 97/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0748 - accuracy: 0.9774 - val_loss: 0.0736 - val_accuracy: 0.9793
Epoch 98/150
283/283 [==============================] - 56s 196ms/step - loss: 0.0603 - accuracy: 0.9818 - val_loss: 0.1096 - val_accuracy: 0.9733
Epoch 99/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0531 - accuracy: 0.9843 - val_loss: 0.0947 - val_accuracy: 0.9753
Epoch 100/150
283/283 [==============================] - 55s 194ms/step - loss: 0.0482 - accuracy: 0.9854 - val_loss: 0.1098 - val_accuracy: 0.9700
Epoch 101/150
283/283 [==============================] - 55s 194ms/step - loss: 0.0467 - accuracy: 0.9862 - val_loss: 0.0929 - val_accuracy: 0.9757
Epoch 102/150
283/283 [==============================] - 55s 194ms/step - loss: 0.0395 - accuracy: 0.9879 - val_loss: 0.1079 - val_accuracy: 0.9717
Epoch 103/150
283/283 [==============================] - 55s 194ms/step - loss: 0.0597 - accuracy: 0.9821 - val_loss: 0.0583 - val_accuracy: 0.9827
Epoch 104/150
283/283 [==============================] - 55s 195ms/step - loss: 0.0368 - accuracy: 0.9884 - val_loss: 0.1852 - val_accuracy: 0.9550
Epoch 105/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0416 - accuracy: 0.9885 - val_loss: 0.1173 - val_accuracy: 0.9707
Epoch 106/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0538 - accuracy: 0.9840 - val_loss: 0.1070 - val_accuracy: 0.9730
Epoch 107/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0609 - accuracy: 0.9824 - val_loss: 0.0792 - val_accuracy: 0.9777
Epoch 108/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0491 - accuracy: 0.9865 - val_loss: 0.0543 - val_accuracy: 0.9870
Epoch 109/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0513 - accuracy: 0.9848 - val_loss: 0.0525 - val_accuracy: 0.9873
Epoch 110/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0371 - accuracy: 0.9890 - val_loss: 0.0707 - val_accuracy: 0.9817
Epoch 111/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0293 - accuracy: 0.9930 - val_loss: 0.2009 - val_accuracy: 0.9417
Epoch 112/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0493 - accuracy: 0.9850 - val_loss: 0.0778 - val_accuracy: 0.9757
Epoch 113/150
283/283 [==============================] - 56s 197ms/step - loss: 0.0407 - accuracy: 0.9890 - val_loss: 0.1273 - val_accuracy: 0.9643
Epoch 114/150
283/283 [==============================] - 58s 202ms/step - loss: 0.0564 - accuracy: 0.9824 - val_loss: 0.0723 - val_accuracy: 0.9810
Epoch 115/150
283/283 [==============================] - 55s 194ms/step - loss: 0.0458 - accuracy: 0.9872 - val_loss: 0.0668 - val_accuracy: 0.9843
Epoch 116/150
283/283 [==============================] - 55s 194ms/step - loss: 0.0603 - accuracy: 0.9818 - val_loss: 0.0892 - val_accuracy: 0.9760
Epoch 117/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0382 - accuracy: 0.9887 - val_loss: 0.0747 - val_accuracy: 0.9800
Epoch 118/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0428 - accuracy: 0.9870 - val_loss: 0.2627 - val_accuracy: 0.9410
Epoch 119/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0285 - accuracy: 0.9911 - val_loss: 0.4958 - val_accuracy: 0.9027
Epoch 120/150
283/283 [==============================] - 56s 196ms/step - loss: 0.0223 - accuracy: 0.9929 - val_loss: 0.0963 - val_accuracy: 0.9790
Epoch 121/150
283/283 [==============================] - 56s 196ms/step - loss: 0.0536 - accuracy: 0.9840 - val_loss: 0.1485 - val_accuracy: 0.9667
Epoch 122/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0504 - accuracy: 0.9870 - val_loss: 0.0860 - val_accuracy: 0.9750
Epoch 123/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0427 - accuracy: 0.9870 - val_loss: 0.0987 - val_accuracy: 0.9753
Epoch 124/150
283/283 [==============================] - 56s 196ms/step - loss: 0.0321 - accuracy: 0.9912 - val_loss: 0.0631 - val_accuracy: 0.9860
Epoch 125/150
283/283 [==============================] - 56s 196ms/step - loss: 0.0337 - accuracy: 0.9909 - val_loss: 0.1287 - val_accuracy: 0.9707
Epoch 126/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0437 - accuracy: 0.9862 - val_loss: 0.1158 - val_accuracy: 0.9747
Epoch 127/150
283/283 [==============================] - 55s 195ms/step - loss: 0.0383 - accuracy: 0.9896 - val_loss: 0.0723 - val_accuracy: 0.9833
Epoch 128/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0324 - accuracy: 0.9903 - val_loss: 0.0689 - val_accuracy: 0.9803
Epoch 129/150
283/283 [==============================] - 55s 194ms/step - loss: 0.0440 - accuracy: 0.9880 - val_loss: 0.0967 - val_accuracy: 0.9777
Epoch 130/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0544 - accuracy: 0.9860 - val_loss: 0.0804 - val_accuracy: 0.9763
Epoch 131/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0477 - accuracy: 0.9863 - val_loss: 0.1692 - val_accuracy: 0.9580
Epoch 132/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0379 - accuracy: 0.9896 - val_loss: 0.0773 - val_accuracy: 0.9833
Epoch 133/150
283/283 [==============================] - 56s 196ms/step - loss: 0.0268 - accuracy: 0.9927 - val_loss: 0.0545 - val_accuracy: 0.9860
Epoch 134/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0476 - accuracy: 0.9878 - val_loss: 0.1215 - val_accuracy: 0.9697
Epoch 135/150
283/283 [==============================] - 56s 196ms/step - loss: 0.0382 - accuracy: 0.9897 - val_loss: 0.0802 - val_accuracy: 0.9843
Epoch 136/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0477 - accuracy: 0.9853 - val_loss: 0.0616 - val_accuracy: 0.9843
Epoch 137/150
283/283 [==============================] - 56s 195ms/step - loss: 0.0408 - accuracy: 0.9877 - val_loss: 0.0611 - val_accuracy: 0.9863
Epoch 138/150
283/283 [==============================] - 56s 196ms/step - loss: 0.0199 - accuracy: 0.9952 - val_loss: 0.1080 - val_accuracy: 0.9780
Epoch 139/150
283/283 [==============================] - 56s 196ms/step - loss: 0.0430 - accuracy: 0.9887 - val_loss: 0.0823 - val_accuracy: 0.9790

history = res128_final_cw_aug.fit(aug_res128_train, validation_data = std_res128_validation, epochs = 150, callbacks = earlyStopper, class_weight = class_weights)
cnnf.model_metric_graph(history, res128_final_cw_aug, std_res31_validation, name = "adapted VGG-16 (with class weights and augmented images applied)", class_names = class_names)

Epoch 1/150
283/283 [==============================] - 52s 176ms/step - loss: 2.5004 - accuracy: 0.1635 - val_loss: 3.5651 - val_accuracy: 0.0773
Epoch 2/150
283/283 [==============================] - 49s 173ms/step - loss: 2.0175 - accuracy: 0.3209 - val_loss: 2.1078 - val_accuracy: 0.2823
Epoch 3/150
283/283 [==============================] - 49s 172ms/step - loss: 1.8039 - accuracy: 0.3949 - val_loss: 3.6867 - val_accuracy: 0.2447
Epoch 4/150
283/283 [==============================] - 49s 171ms/step - loss: 1.7024 - accuracy: 0.4342 - val_loss: 3.4413 - val_accuracy: 0.2903
Epoch 5/150
283/283 [==============================] - 49s 172ms/step - loss: 1.5026 - accuracy: 0.5006 - val_loss: 1.5824 - val_accuracy: 0.4677
Epoch 6/150
283/283 [==============================] - 49s 171ms/step - loss: 1.3714 - accuracy: 0.5439 - val_loss: 2.2896 - val_accuracy: 0.3827
Epoch 7/150
283/283 [==============================] - 49s 172ms/step - loss: 1.2926 - accuracy: 0.5840 - val_loss: 1.4818 - val_accuracy: 0.5037
Epoch 8/150
283/283 [==============================] - 49s 172ms/step - loss: 1.1617 - accuracy: 0.6238 - val_loss: 1.3014 - val_accuracy: 0.6030
Epoch 9/150
283/283 [==============================] - 49s 171ms/step - loss: 1.0591 - accuracy: 0.6540 - val_loss: 1.5194 - val_accuracy: 0.5337
Epoch 10/150
283/283 [==============================] - 49s 170ms/step - loss: 0.9973 - accuracy: 0.6708 - val_loss: 1.7059 - val_accuracy: 0.5457
Epoch 11/150
283/283 [==============================] - 49s 172ms/step - loss: 0.9342 - accuracy: 0.6938 - val_loss: 1.2007 - val_accuracy: 0.6120
Epoch 12/150
283/283 [==============================] - 49s 171ms/step - loss: 0.8907 - accuracy: 0.7111 - val_loss: 1.3605 - val_accuracy: 0.5957
Epoch 13/150
283/283 [==============================] - 49s 173ms/step - loss: 0.8066 - accuracy: 0.7337 - val_loss: 1.6680 - val_accuracy: 0.6007
Epoch 14/150
283/283 [==============================] - 49s 173ms/step - loss: 0.7692 - accuracy: 0.7493 - val_loss: 1.1905 - val_accuracy: 0.6353
Epoch 15/150
283/283 [==============================] - 50s 174ms/step - loss: 0.6997 - accuracy: 0.7696 - val_loss: 1.1700 - val_accuracy: 0.6907
Epoch 16/150
283/283 [==============================] - 50s 174ms/step - loss: 0.6801 - accuracy: 0.7806 - val_loss: 0.9963 - val_accuracy: 0.6943
Epoch 17/150
283/283 [==============================] - 49s 173ms/step - loss: 0.6016 - accuracy: 0.8035 - val_loss: 1.3738 - val_accuracy: 0.6400
Epoch 18/150
283/283 [==============================] - 49s 172ms/step - loss: 0.6011 - accuracy: 0.8027 - val_loss: 0.9628 - val_accuracy: 0.7353
Epoch 19/150
283/283 [==============================] - 49s 171ms/step - loss: 0.5178 - accuracy: 0.8365 - val_loss: 0.9411 - val_accuracy: 0.7120
Epoch 20/150
283/283 [==============================] - 50s 174ms/step - loss: 0.5034 - accuracy: 0.8383 - val_loss: 0.5420 - val_accuracy: 0.8273
Epoch 21/150
283/283 [==============================] - 49s 171ms/step - loss: 0.4621 - accuracy: 0.8562 - val_loss: 0.6760 - val_accuracy: 0.8007
Epoch 22/150
283/283 [==============================] - 48s 169ms/step - loss: 0.4097 - accuracy: 0.8746 - val_loss: 0.6002 - val_accuracy: 0.8207
Epoch 23/150
283/283 [==============================] - 49s 173ms/step - loss: 0.3680 - accuracy: 0.8818 - val_loss: 0.5455 - val_accuracy: 0.8387
Epoch 24/150
283/283 [==============================] - 49s 173ms/step - loss: 0.3573 - accuracy: 0.8867 - val_loss: 0.3993 - val_accuracy: 0.8833
Epoch 25/150
283/283 [==============================] - 50s 175ms/step - loss: 0.3461 - accuracy: 0.8907 - val_loss: 0.3311 - val_accuracy: 0.8970
Epoch 26/150
283/283 [==============================] - 49s 172ms/step - loss: 0.3201 - accuracy: 0.8958 - val_loss: 0.3173 - val_accuracy: 0.9050
Epoch 27/150
283/283 [==============================] - 48s 170ms/step - loss: 0.2840 - accuracy: 0.9071 - val_loss: 0.2990 - val_accuracy: 0.9093
Epoch 28/150
283/283 [==============================] - 49s 172ms/step - loss: 0.2946 - accuracy: 0.9078 - val_loss: 0.9116 - val_accuracy: 0.7397
Epoch 29/150
283/283 [==============================] - 49s 171ms/step - loss: 0.3083 - accuracy: 0.9021 - val_loss: 0.3049 - val_accuracy: 0.9083
Epoch 30/150
283/283 [==============================] - 49s 171ms/step - loss: 0.2673 - accuracy: 0.9157 - val_loss: 0.2936 - val_accuracy: 0.9123
Epoch 31/150
283/283 [==============================] - 48s 170ms/step - loss: 0.2574 - accuracy: 0.9171 - val_loss: 0.3694 - val_accuracy: 0.8930
Epoch 32/150
283/283 [==============================] - 49s 172ms/step - loss: 0.2057 - accuracy: 0.9329 - val_loss: 0.4277 - val_accuracy: 0.8797
Epoch 33/150
283/283 [==============================] - 49s 173ms/step - loss: 0.2190 - accuracy: 0.9302 - val_loss: 0.3843 - val_accuracy: 0.8887
Epoch 34/150
283/283 [==============================] - 49s 171ms/step - loss: 0.2292 - accuracy: 0.9269 - val_loss: 0.5753 - val_accuracy: 0.8563
Epoch 35/150
283/283 [==============================] - 49s 171ms/step - loss: 0.2119 - accuracy: 0.9337 - val_loss: 0.2298 - val_accuracy: 0.9367
Epoch 36/150
283/283 [==============================] - 49s 172ms/step - loss: 0.2260 - accuracy: 0.9307 - val_loss: 0.2862 - val_accuracy: 0.9147
Epoch 37/150
283/283 [==============================] - 48s 169ms/step - loss: 0.1793 - accuracy: 0.9405 - val_loss: 0.2175 - val_accuracy: 0.9370
Epoch 38/150
283/283 [==============================] - 49s 173ms/step - loss: 0.1777 - accuracy: 0.9451 - val_loss: 1.2564 - val_accuracy: 0.7143
Epoch 39/150
283/283 [==============================] - 48s 169ms/step - loss: 0.1907 - accuracy: 0.9413 - val_loss: 0.2530 - val_accuracy: 0.9307
Epoch 40/150
283/283 [==============================] - 49s 171ms/step - loss: 0.1757 - accuracy: 0.9448 - val_loss: 0.4177 - val_accuracy: 0.8863
Epoch 41/150
283/283 [==============================] - 49s 172ms/step - loss: 0.1819 - accuracy: 0.9414 - val_loss: 0.5381 - val_accuracy: 0.8623
Epoch 42/150
283/283 [==============================] - 49s 170ms/step - loss: 0.2285 - accuracy: 0.9258 - val_loss: 0.3583 - val_accuracy: 0.8967
Epoch 43/150
283/283 [==============================] - 48s 170ms/step - loss: 0.1477 - accuracy: 0.9513 - val_loss: 0.5804 - val_accuracy: 0.8577
Epoch 44/150
283/283 [==============================] - 49s 173ms/step - loss: 0.1469 - accuracy: 0.9525 - val_loss: 0.2819 - val_accuracy: 0.9317
Epoch 45/150
283/283 [==============================] - 49s 170ms/step - loss: 0.1519 - accuracy: 0.9508 - val_loss: 0.3036 - val_accuracy: 0.9123
Epoch 46/150
283/283 [==============================] - 50s 173ms/step - loss: 0.1352 - accuracy: 0.9541 - val_loss: 0.2183 - val_accuracy: 0.9433
Epoch 47/150
283/283 [==============================] - 50s 174ms/step - loss: 0.1521 - accuracy: 0.9545 - val_loss: 0.3952 - val_accuracy: 0.8960
Epoch 48/150
283/283 [==============================] - 49s 172ms/step - loss: 0.1192 - accuracy: 0.9620 - val_loss: 0.1891 - val_accuracy: 0.9483
Epoch 49/150
283/283 [==============================] - 48s 170ms/step - loss: 0.1454 - accuracy: 0.9530 - val_loss: 0.1675 - val_accuracy: 0.9453
Epoch 50/150
283/283 [==============================] - 49s 173ms/step - loss: 0.1351 - accuracy: 0.9603 - val_loss: 0.5026 - val_accuracy: 0.8723
Epoch 51/150
283/283 [==============================] - 49s 171ms/step - loss: 0.1417 - accuracy: 0.9530 - val_loss: 0.9514 - val_accuracy: 0.7700
Epoch 52/150
283/283 [==============================] - 49s 171ms/step - loss: 0.2366 - accuracy: 0.9260 - val_loss: 0.2037 - val_accuracy: 0.9413
Epoch 53/150
283/283 [==============================] - 49s 173ms/step - loss: 0.1590 - accuracy: 0.9488 - val_loss: 0.1885 - val_accuracy: 0.9443
Epoch 54/150
283/283 [==============================] - 49s 172ms/step - loss: 0.1331 - accuracy: 0.9591 - val_loss: 0.1361 - val_accuracy: 0.9580
Epoch 55/150
283/283 [==============================] - 49s 171ms/step - loss: 0.1074 - accuracy: 0.9650 - val_loss: 0.3001 - val_accuracy: 0.9180
Epoch 56/150
283/283 [==============================] - 49s 172ms/step - loss: 0.1461 - accuracy: 0.9568 - val_loss: 0.1865 - val_accuracy: 0.9497
Epoch 57/150
283/283 [==============================] - 49s 173ms/step - loss: 0.1090 - accuracy: 0.9640 - val_loss: 0.1908 - val_accuracy: 0.9523
Epoch 58/150
283/283 [==============================] - 49s 173ms/step - loss: 0.1238 - accuracy: 0.9639 - val_loss: 0.1664 - val_accuracy: 0.9543
Epoch 59/150
283/283 [==============================] - 49s 173ms/step - loss: 0.0975 - accuracy: 0.9671 - val_loss: 0.4894 - val_accuracy: 0.8770
Epoch 60/150
283/283 [==============================] - 50s 173ms/step - loss: 0.1392 - accuracy: 0.9543 - val_loss: 0.2647 - val_accuracy: 0.9270
Epoch 61/150
283/283 [==============================] - 49s 172ms/step - loss: 0.0937 - accuracy: 0.9694 - val_loss: 0.1482 - val_accuracy: 0.9590
Epoch 62/150
283/283 [==============================] - 49s 171ms/step - loss: 0.1195 - accuracy: 0.9651 - val_loss: 0.3310 - val_accuracy: 0.9110
Epoch 63/150
283/283 [==============================] - 49s 173ms/step - loss: 0.0957 - accuracy: 0.9722 - val_loss: 0.1399 - val_accuracy: 0.9640
Epoch 64/150
283/283 [==============================] - 49s 172ms/step - loss: 0.0808 - accuracy: 0.9724 - val_loss: 0.2927 - val_accuracy: 0.9243
Epoch 65/150
283/283 [==============================] - 49s 172ms/step - loss: 0.0876 - accuracy: 0.9721 - val_loss: 0.1358 - val_accuracy: 0.9650
Epoch 66/150
283/283 [==============================] - 49s 172ms/step - loss: 0.0887 - accuracy: 0.9723 - val_loss: 0.6669 - val_accuracy: 0.8390
Epoch 67/150
283/283 [==============================] - 49s 170ms/step - loss: 0.0887 - accuracy: 0.9719 - val_loss: 0.1787 - val_accuracy: 0.9533
Epoch 68/150
283/283 [==============================] - 50s 173ms/step - loss: 0.1057 - accuracy: 0.9677 - val_loss: 0.1420 - val_accuracy: 0.9587
Epoch 69/150
283/283 [==============================] - 49s 171ms/step - loss: 0.0741 - accuracy: 0.9736 - val_loss: 0.1155 - val_accuracy: 0.9723
Epoch 70/150
283/283 [==============================] - 49s 171ms/step - loss: 0.0923 - accuracy: 0.9704 - val_loss: 0.5099 - val_accuracy: 0.9107
Epoch 71/150
283/283 [==============================] - 50s 174ms/step - loss: 0.0895 - accuracy: 0.9722 - val_loss: 0.5233 - val_accuracy: 0.8720
Epoch 72/150
283/283 [==============================] - 48s 169ms/step - loss: 0.0836 - accuracy: 0.9721 - val_loss: 0.5088 - val_accuracy: 0.8873
Epoch 73/150
283/283 [==============================] - 48s 167ms/step - loss: 0.1560 - accuracy: 0.9520 - val_loss: 0.1200 - val_accuracy: 0.9683
Epoch 74/150
283/283 [==============================] - 48s 167ms/step - loss: 0.0851 - accuracy: 0.9716 - val_loss: 0.2701 - val_accuracy: 0.9297
Epoch 75/150
283/283 [==============================] - 50s 175ms/step - loss: 0.0894 - accuracy: 0.9699 - val_loss: 0.2315 - val_accuracy: 0.9337
Epoch 76/150
283/283 [==============================] - 50s 174ms/step - loss: 0.0675 - accuracy: 0.9782 - val_loss: 0.0829 - val_accuracy: 0.9793
Epoch 77/150
283/283 [==============================] - 49s 173ms/step - loss: 0.0490 - accuracy: 0.9828 - val_loss: 0.1383 - val_accuracy: 0.9643
Epoch 78/150
283/283 [==============================] - 50s 174ms/step - loss: 0.0707 - accuracy: 0.9773 - val_loss: 0.1208 - val_accuracy: 0.9690
Epoch 79/150
283/283 [==============================] - 50s 174ms/step - loss: 0.0727 - accuracy: 0.9773 - val_loss: 0.1951 - val_accuracy: 0.9523
Epoch 80/150
283/283 [==============================] - 50s 175ms/step - loss: 0.0627 - accuracy: 0.9780 - val_loss: 0.1535 - val_accuracy: 0.9633
Epoch 81/150
283/283 [==============================] - 49s 170ms/step - loss: 0.0950 - accuracy: 0.9704 - val_loss: 0.1080 - val_accuracy: 0.9727
Epoch 82/150
283/283 [==============================] - 48s 168ms/step - loss: 0.0913 - accuracy: 0.9700 - val_loss: 0.2461 - val_accuracy: 0.9393
Epoch 83/150
283/283 [==============================] - 48s 168ms/step - loss: 0.0634 - accuracy: 0.9791 - val_loss: 0.0867 - val_accuracy: 0.9787
Epoch 84/150
283/283 [==============================] - 48s 170ms/step - loss: 0.0825 - accuracy: 0.9749 - val_loss: 0.1481 - val_accuracy: 0.9593
Epoch 85/150
283/283 [==============================] - 48s 168ms/step - loss: 0.0564 - accuracy: 0.9823 - val_loss: 0.1245 - val_accuracy: 0.9657
Epoch 86/150
283/283 [==============================] - 48s 169ms/step - loss: 0.0611 - accuracy: 0.9808 - val_loss: 0.0954 - val_accuracy: 0.9747
Epoch 87/150
283/283 [==============================] - 48s 169ms/step - loss: 0.1055 - accuracy: 0.9690 - val_loss: 0.1596 - val_accuracy: 0.9557
Epoch 88/150
283/283 [==============================] - 49s 171ms/step - loss: 0.0732 - accuracy: 0.9785 - val_loss: 0.0733 - val_accuracy: 0.9803
Epoch 89/150
283/283 [==============================] - 49s 171ms/step - loss: 0.0508 - accuracy: 0.9831 - val_loss: 0.1036 - val_accuracy: 0.9750
Epoch 90/150
283/283 [==============================] - 49s 170ms/step - loss: 0.0527 - accuracy: 0.9840 - val_loss: 0.1038 - val_accuracy: 0.9697
Epoch 91/150
283/283 [==============================] - 48s 169ms/step - loss: 0.0605 - accuracy: 0.9811 - val_loss: 0.2352 - val_accuracy: 0.9373
Epoch 92/150
283/283 [==============================] - 48s 169ms/step - loss: 0.0784 - accuracy: 0.9750 - val_loss: 0.3550 - val_accuracy: 0.9037
Epoch 93/150
283/283 [==============================] - 48s 169ms/step - loss: 0.0585 - accuracy: 0.9808 - val_loss: 0.1337 - val_accuracy: 0.9663
Epoch 94/150
283/283 [==============================] - 48s 169ms/step - loss: 0.0724 - accuracy: 0.9768 - val_loss: 0.0942 - val_accuracy: 0.9750
Epoch 95/150
283/283 [==============================] - 48s 168ms/step - loss: 0.0485 - accuracy: 0.9856 - val_loss: 0.2701 - val_accuracy: 0.9377
Epoch 96/150
283/283 [==============================] - 49s 171ms/step - loss: 0.0821 - accuracy: 0.9757 - val_loss: 1.3307 - val_accuracy: 0.7247
Epoch 97/150
283/283 [==============================] - 49s 170ms/step - loss: 0.0636 - accuracy: 0.9786 - val_loss: 0.1075 - val_accuracy: 0.9710
Epoch 98/150
283/283 [==============================] - 48s 169ms/step - loss: 0.0579 - accuracy: 0.9821 - val_loss: 0.0937 - val_accuracy: 0.9757
Epoch 99/150
283/283 [==============================] - 49s 170ms/step - loss: 0.0393 - accuracy: 0.9858 - val_loss: 0.0910 - val_accuracy: 0.9787
Epoch 100/150
283/283 [==============================] - 48s 170ms/step - loss: 0.0459 - accuracy: 0.9852 - val_loss: 0.0753 - val_accuracy: 0.9840
Epoch 101/150
283/283 [==============================] - 48s 170ms/step - loss: 0.0820 - accuracy: 0.9755 - val_loss: 0.1075 - val_accuracy: 0.9750
Epoch 102/150
283/283 [==============================] - 49s 170ms/step - loss: 0.0452 - accuracy: 0.9877 - val_loss: 0.0922 - val_accuracy: 0.9763
Epoch 103/150
283/283 [==============================] - 48s 169ms/step - loss: 0.0585 - accuracy: 0.9798 - val_loss: 0.1073 - val_accuracy: 0.9713
Epoch 104/150
283/283 [==============================] - 49s 170ms/step - loss: 0.0591 - accuracy: 0.9811 - val_loss: 0.1308 - val_accuracy: 0.9657
Epoch 105/150
283/283 [==============================] - 48s 167ms/step - loss: 0.0823 - accuracy: 0.9751 - val_loss: 0.1054 - val_accuracy: 0.9763
Epoch 106/150
283/283 [==============================] - 49s 170ms/step - loss: 0.0413 - accuracy: 0.9863 - val_loss: 0.0891 - val_accuracy: 0.9767
Epoch 107/150
283/283 [==============================] - 48s 168ms/step - loss: 0.0480 - accuracy: 0.9844 - val_loss: 0.0964 - val_accuracy: 0.9793
Epoch 108/150
283/283 [==============================] - 49s 170ms/step - loss: 0.0680 - accuracy: 0.9792 - val_loss: 0.1512 - val_accuracy: 0.9617
Epoch 109/150
283/283 [==============================] - 49s 171ms/step - loss: 0.0611 - accuracy: 0.9817 - val_loss: 0.0663 - val_accuracy: 0.9833
Epoch 110/150
283/283 [==============================] - 49s 170ms/step - loss: 0.0416 - accuracy: 0.9864 - val_loss: 0.1113 - val_accuracy: 0.9750
Epoch 111/150
283/283 [==============================] - 48s 169ms/step - loss: 0.0486 - accuracy: 0.9854 - val_loss: 0.1975 - val_accuracy: 0.9517
Epoch 112/150
283/283 [==============================] - 50s 174ms/step - loss: 0.0886 - accuracy: 0.9722 - val_loss: 0.0933 - val_accuracy: 0.9750
Epoch 113/150
283/283 [==============================] - 48s 169ms/step - loss: 0.0423 - accuracy: 0.9867 - val_loss: 0.0714 - val_accuracy: 0.9797
Epoch 114/150
283/283 [==============================] - 49s 170ms/step - loss: 0.0570 - accuracy: 0.9822 - val_loss: 0.0959 - val_accuracy: 0.9750
Epoch 115/150
283/283 [==============================] - 48s 170ms/step - loss: 0.0528 - accuracy: 0.9847 - val_loss: 0.0688 - val_accuracy: 0.9823
Epoch 116/150
283/283 [==============================] - 48s 168ms/step - loss: 0.0354 - accuracy: 0.9887 - val_loss: 0.0891 - val_accuracy: 0.9773
Epoch 117/150
283/283 [==============================] - 48s 169ms/step - loss: 0.0858 - accuracy: 0.9730 - val_loss: 0.1173 - val_accuracy: 0.9683
Epoch 118/150
283/283 [==============================] - 48s 168ms/step - loss: 0.0475 - accuracy: 0.9857 - val_loss: 0.0828 - val_accuracy: 0.9813
Epoch 119/150
283/283 [==============================] - 49s 170ms/step - loss: 0.0361 - accuracy: 0.9870 - val_loss: 0.1099 - val_accuracy: 0.9687
Epoch 120/150
283/283 [==============================] - 49s 170ms/step - loss: 0.0400 - accuracy: 0.9883 - val_loss: 0.1261 - val_accuracy: 0.9663
Epoch 121/150
283/283 [==============================] - 48s 169ms/step - loss: 0.0568 - accuracy: 0.9807 - val_loss: 0.0760 - val_accuracy: 0.9830
Epoch 122/150
283/283 [==============================] - 49s 171ms/step - loss: 0.0506 - accuracy: 0.9846 - val_loss: 0.1194 - val_accuracy: 0.9717
Epoch 123/150
283/283 [==============================] - 48s 169ms/step - loss: 0.0507 - accuracy: 0.9840 - val_loss: 0.0665 - val_accuracy: 0.9840
Epoch 124/150
283/283 [==============================] - 49s 171ms/step - loss: 0.0429 - accuracy: 0.9878 - val_loss: 0.0820 - val_accuracy: 0.9820
Epoch 125/150
283/283 [==============================] - 48s 170ms/step - loss: 0.0210 - accuracy: 0.9925 - val_loss: 0.0724 - val_accuracy: 0.9833
Epoch 126/150
283/283 [==============================] - 48s 168ms/step - loss: 0.0496 - accuracy: 0.9845 - val_loss: 0.0931 - val_accuracy: 0.9753
Epoch 127/150
283/283 [==============================] - 49s 170ms/step - loss: 0.0582 - accuracy: 0.9842 - val_loss: 0.0891 - val_accuracy: 0.9750
Epoch 128/150
283/283 [==============================] - 49s 170ms/step - loss: 0.0486 - accuracy: 0.9855 - val_loss: 0.1856 - val_accuracy: 0.9627
Epoch 129/150
283/283 [==============================] - 49s 171ms/step - loss: 0.0380 - accuracy: 0.9875 - val_loss: 0.1100 - val_accuracy: 0.9723
Epoch 130/150
283/283 [==============================] - 49s 170ms/step - loss: 0.0386 - accuracy: 0.9896 - val_loss: 0.1057 - val_accuracy: 0.9700
Epoch 131/150
283/283 [==============================] - 48s 169ms/step - loss: 0.0577 - accuracy: 0.9829 - val_loss: 0.1242 - val_accuracy: 0.9713
Epoch 132/150
283/283 [==============================] - 49s 170ms/step - loss: 0.0446 - accuracy: 0.9868 - val_loss: 0.0760 - val_accuracy: 0.9820
Epoch 133/150
283/283 [==============================] - 49s 170ms/step - loss: 0.0368 - accuracy: 0.9880 - val_loss: 0.0924 - val_accuracy: 0.9807
Epoch 134/150
283/283 [==============================] - 49s 171ms/step - loss: 0.0438 - accuracy: 0.9852 - val_loss: 0.0673 - val_accuracy: 0.9833
Epoch 135/150
283/283 [==============================] - 49s 171ms/step - loss: 0.0484 - accuracy: 0.9845 - val_loss: 0.0857 - val_accuracy: 0.9803
Epoch 136/150
283/283 [==============================] - 49s 171ms/step - loss: 0.0457 - accuracy: 0.9877 - val_loss: 0.1088 - val_accuracy: 0.9723
Epoch 137/150
283/283 [==============================] - 48s 170ms/step - loss: 0.0333 - accuracy: 0.9897 - val_loss: 0.0834 - val_accuracy: 0.9737
Epoch 138/150
283/283 [==============================] - 49s 171ms/step - loss: 0.0489 - accuracy: 0.9842 - val_loss: 0.1568 - val_accuracy: 0.9573
Epoch 139/150
283/283 [==============================] - 48s 170ms/step - loss: 0.0349 - accuracy: 0.9886 - val_loss: 0.1532 - val_accuracy: 0.9620

improvement128_model_list = [res128_vggmodel, res128_final_cw, res128_final_aug, res128_final_cw_aug]
improvement128_data_list = [std_res128_validation, std_res128_validation, std_res128_validation, std_res128_validation]
name_list = ["VGG Model with no improvements", "Class Weights Applied", "Augmentation Applied", "Class Weights and Augmentation Applied"]
cnnf.model_comparison(improvement128_model_list, improvement128_data_list, name_list)

94/94 [==============================] - 5s 42ms/step - loss: 0.1726 - accuracy: 0.9653
94/94 [==============================] - 5s 45ms/step - loss: 0.2479 - accuracy: 0.9527
94/94 [==============================] - 4s 41ms/step - loss: 0.0823 - accuracy: 0.9790
94/94 [==============================] - 4s 41ms/step - loss: 0.1532 - accuracy: 0.9620

From the above, we can see that all of the improvements that we attempted have had a improvement on the model's performance, with the application of Augmentation only having the most profound improvement, with accuracy and loss both being the best performing among all the 4 permutations. Hence, we shall now select this model and hypertune it. One interesting observation is that appling class weights actually causes the model to perform worse, with a 43% drop in loss performance.

From this, we can conclude that augmentation is useful for the 128 x 128 image input size.

First, we shall save the model that we want to hypertune, so that we can hypertune seperately, saving on computational resources.

res128_final_aug.save('res128_best.h5')

---

Model Hypertuning (31 x 31)

---

Now that we have completed the larger steps in model improvement, we shall now take the smaller, final step to allow the model to reach it's peak performance. We shall do this by hypertuning the parameters of the model. For our case, we shall only tune the optimizer, as the loss function is already very well suited for our use case. We shall tune the models using the Adam, SGD, and RMSProp optimizers.

Stochastic Gradient Descent (SGD) is one of the most basic and widely used optimization algorithms in machine learning and deep learning. It's a variant of gradient descent where instead of performing computations on the whole dataset – which can be computationally intensive for large datasets – SGD updates the model's weights using only a single or a few samples at a time. This makes the algorithm much faster and more suitable for large datasets. The formula for SGD is as follows. $$ w_{t+1} = w_t - \eta \cdot \nabla L(w_t, x_i, y_i) $$

\begin{aligned} \text{where} \\ w_{t+1} &\text{ is the updated weight vector at time } t+1, \\ w_t &\text{ is the weight vector at time } t, \\ \eta &\text{ is the learning rate, and} \\ \nabla L(w_t, x_i, y_i) &\text{ is the gradient of the loss function } L \text{ with respect to the weights } w, \text{ evaluated at a randomly chosen data point } (x_i, y_i). \end{aligned}

RMSProp, short for Root Mean Square Propagation, is an adaptive learning rate method proposed by Geoffrey Hinton. It addresses some of the limitations of SGD, especially in the context of minimizing functions in very high-dimensional spaces. RMSProp adjusts the learning rate for each weight based on the recent magnitudes of the gradients for that weight. This means that the learning rate is reduced for weights that consistently receive high gradients, which helps in faster convergence especially in situations involving oscillations. The formula for RMSProp is as follows.

\begin{aligned} v_{t+1} &= \beta \cdot v_t + (1 - \beta) \cdot (\nabla L(w_t))^2 \\ w_{t+1} &= w_t - \frac{\eta}{\sqrt{v_{t+1} + \epsilon}} \cdot \nabla L(w_t) \\ \text{where} \\ v_{t+1} &\text{ is the exponentially decaying average of squared gradients} \\ w_{t+1} &\text{ is the updated weight vector at time } t+1 \\ w_t &\text{ is the weight vector at time } t \\ \eta &\text{ is the learning rate} \\ \beta &\text{ is the decay rate, controlling the moving average of squared gradients} \\ \epsilon &\text{ is a small number to prevent division by zero} \\ \nabla L(w_t) &\text{ is the gradient of the loss function } L \text{ with respect to the weights } w \text{ at time } t \end{aligned}

Despite Adam and RMSProp being both able to adjust their learning rates, the values which they start at are still quite important to how the model performs. Hence, we shall do 3 permutations for each optimizer, and see how they improve from there.

earlyStopper = EarlyStopping(monitor = 'val_loss', patience = 15)

def model_hypertuner(optimizer_list, name_list, model_list):
    for optimizer, name, model in zip(optimizer_list, name_list, model_list):
        clear_output(wait=True)
        print(f"Now attempting to tune {name}")
        model.compile(optimizer = optimizer, loss = "sparse_categorical_crossentropy", metrics=['accuracy'])
        model.fit(aug_res31_train, validation_data = std_res31_validation, epochs = 75, callbacks = earlyStopper, class_weight = class_weights)
        gc.collect()

from tensorflow.keras.optimizers import Adam
from tensorflow.keras.optimizers import SGD
from tensorflow.keras.optimizers import RMSprop
optimizer_list = [
    Adam(),
    Adam(learning_rate=0.005),
    Adam(learning_rate=0.0002),
    SGD(learning_rate=0.005),
    SGD(learning_rate=0.05),
    SGD(),
    RMSprop(learning_rate=0.05),
    RMSprop(learning_rate=0.002),
    RMSprop(),
]

name_list = [
    'Adam_LR_0.001',
    'Adam_LR_0.005',
    'Adam_LR_0.0002',
    'SGD_LR_0.005',
    'SGD_LR_0.05',
    'SGD_LR_0.01',
    'RMSprop_LR_0.05',
    'RMSprop_LR_0.002',
    'RMSprop_LR_0.01'
]

tuned_31_Adam_LR_0_001 = load_model('res31_best.h5')
tuned_31_Adam_LR_0_005 = load_model('res31_best.h5')
tuned_31_Adam_LR_0_0002 = load_model('res31_best.h5')
tuned_31_SGD_LR_0_005 = load_model('res31_best.h5')
tuned_31_SGD_LR_0_05 = load_model('res31_best.h5')
tuned_31_SGD_LR_0_01 = load_model('res31_best.h5')
tuned_31_RMSprop_LR_0_05 = load_model('res31_best.h5')
tuned_31_RMSprop_LR_0_002 = load_model('res31_best.h5')
tuned_31_RMSprop_LR_0_01 = load_model('res31_best.h5')


output_list = [tuned_31_Adam_LR_0_001, tuned_31_Adam_LR_0_005, tuned_31_Adam_LR_0_0002, tuned_31_SGD_LR_0_005, tuned_31_SGD_LR_0_05, tuned_31_SGD_LR_0_01, tuned_31_RMSprop_LR_0_05, tuned_31_RMSprop_LR_0_002, tuned_31_RMSprop_LR_0_01]

data_list = [std_res31_validation, std_res31_validation, std_res31_validation, std_res31_validation, std_res31_validation, std_res31_validation, std_res31_validation, std_res31_validation, std_res31_validation]

model_hypertuner(optimizer_list, name_list, output_list)
cnnf.model_comparison(output_list, data_list, name_list)

Now attempting to tune RMSprop_LR_0.01
Epoch 1/75
283/283 [==============================] - 15s 43ms/step - loss: 0.3753 - accuracy: 0.8856 - val_loss: 1.4580 - val_accuracy: 0.6820
Epoch 2/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2982 - accuracy: 0.9042 - val_loss: 1.0157 - val_accuracy: 0.7587
Epoch 3/75
283/283 [==============================] - 12s 43ms/step - loss: 0.2779 - accuracy: 0.9160 - val_loss: 1.1390 - val_accuracy: 0.7510
Epoch 4/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2681 - accuracy: 0.9205 - val_loss: 1.0231 - val_accuracy: 0.7717
Epoch 5/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2714 - accuracy: 0.9167 - val_loss: 0.6141 - val_accuracy: 0.8437
Epoch 6/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2758 - accuracy: 0.9198 - val_loss: 0.9498 - val_accuracy: 0.7987
Epoch 7/75
283/283 [==============================] - 12s 43ms/step - loss: 0.2674 - accuracy: 0.9187 - val_loss: 0.9154 - val_accuracy: 0.7980
Epoch 8/75
283/283 [==============================] - 12s 43ms/step - loss: 0.2611 - accuracy: 0.9220 - val_loss: 0.7599 - val_accuracy: 0.8590
Epoch 9/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2532 - accuracy: 0.9243 - val_loss: 0.8581 - val_accuracy: 0.8103
Epoch 10/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2465 - accuracy: 0.9270 - val_loss: 0.7018 - val_accuracy: 0.8480
Epoch 11/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2580 - accuracy: 0.9241 - val_loss: 0.6722 - val_accuracy: 0.8483
Epoch 12/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2386 - accuracy: 0.9294 - val_loss: 0.5131 - val_accuracy: 0.8807
Epoch 13/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2631 - accuracy: 0.9278 - val_loss: 0.6090 - val_accuracy: 0.8503
Epoch 14/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2611 - accuracy: 0.9279 - val_loss: 0.7857 - val_accuracy: 0.8163
Epoch 15/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2399 - accuracy: 0.9300 - val_loss: 0.5388 - val_accuracy: 0.8897
Epoch 16/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2230 - accuracy: 0.9330 - val_loss: 0.7800 - val_accuracy: 0.8423
Epoch 17/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2522 - accuracy: 0.9303 - val_loss: 0.5498 - val_accuracy: 0.8553
Epoch 18/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2564 - accuracy: 0.9255 - val_loss: 0.5863 - val_accuracy: 0.8690
Epoch 19/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2259 - accuracy: 0.9329 - val_loss: 0.5460 - val_accuracy: 0.8667
Epoch 20/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2411 - accuracy: 0.9288 - val_loss: 0.5891 - val_accuracy: 0.8613
Epoch 21/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2255 - accuracy: 0.9323 - val_loss: 0.8443 - val_accuracy: 0.7940
Epoch 22/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2379 - accuracy: 0.9308 - val_loss: 0.3707 - val_accuracy: 0.9207
Epoch 23/75
283/283 [==============================] - 12s 43ms/step - loss: 0.2248 - accuracy: 0.9370 - val_loss: 0.5791 - val_accuracy: 0.8963
Epoch 24/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2338 - accuracy: 0.9329 - val_loss: 0.5993 - val_accuracy: 0.8737
Epoch 25/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2336 - accuracy: 0.9327 - val_loss: 0.6589 - val_accuracy: 0.8590
Epoch 26/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2347 - accuracy: 0.9363 - val_loss: 0.6520 - val_accuracy: 0.8697
Epoch 27/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2266 - accuracy: 0.9318 - val_loss: 1.0028 - val_accuracy: 0.8110
Epoch 28/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2390 - accuracy: 0.9323 - val_loss: 0.7635 - val_accuracy: 0.8417
Epoch 29/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2141 - accuracy: 0.9368 - val_loss: 0.6427 - val_accuracy: 0.8703
Epoch 30/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2205 - accuracy: 0.9399 - val_loss: 0.9982 - val_accuracy: 0.8330
Epoch 31/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2116 - accuracy: 0.9390 - val_loss: 0.4486 - val_accuracy: 0.9077
Epoch 32/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2273 - accuracy: 0.9362 - val_loss: 0.5304 - val_accuracy: 0.8970
Epoch 33/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2276 - accuracy: 0.9364 - val_loss: 0.3944 - val_accuracy: 0.9010
Epoch 34/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2420 - accuracy: 0.9361 - val_loss: 0.6279 - val_accuracy: 0.8727
Epoch 35/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2139 - accuracy: 0.9418 - val_loss: 0.3579 - val_accuracy: 0.9113
Epoch 36/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2133 - accuracy: 0.9399 - val_loss: 0.5748 - val_accuracy: 0.8863
Epoch 37/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2149 - accuracy: 0.9413 - val_loss: 0.6420 - val_accuracy: 0.8863
Epoch 38/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2268 - accuracy: 0.9403 - val_loss: 0.5430 - val_accuracy: 0.8957
Epoch 39/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2043 - accuracy: 0.9446 - val_loss: 0.3671 - val_accuracy: 0.9177
Epoch 40/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2241 - accuracy: 0.9426 - val_loss: 0.4018 - val_accuracy: 0.9143
Epoch 41/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2272 - accuracy: 0.9403 - val_loss: 1.0013 - val_accuracy: 0.7870
Epoch 42/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2155 - accuracy: 0.9407 - val_loss: 0.5323 - val_accuracy: 0.8800
Epoch 43/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2149 - accuracy: 0.9423 - val_loss: 0.3520 - val_accuracy: 0.9277
Epoch 44/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2213 - accuracy: 0.9413 - val_loss: 0.7185 - val_accuracy: 0.8837
Epoch 45/75
283/283 [==============================] - 12s 43ms/step - loss: 0.2394 - accuracy: 0.9394 - val_loss: 1.1354 - val_accuracy: 0.8167
Epoch 46/75
283/283 [==============================] - 13s 45ms/step - loss: 0.2061 - accuracy: 0.9442 - val_loss: 0.6748 - val_accuracy: 0.8583
Epoch 47/75
283/283 [==============================] - 14s 47ms/step - loss: 0.2180 - accuracy: 0.9432 - val_loss: 0.5281 - val_accuracy: 0.8820
Epoch 48/75
283/283 [==============================] - 14s 47ms/step - loss: 0.2106 - accuracy: 0.9425 - val_loss: 0.5566 - val_accuracy: 0.8950
Epoch 49/75
283/283 [==============================] - 14s 47ms/step - loss: 0.2249 - accuracy: 0.9415 - val_loss: 0.8661 - val_accuracy: 0.8370
Epoch 50/75
283/283 [==============================] - 14s 47ms/step - loss: 0.2333 - accuracy: 0.9442 - val_loss: 0.5835 - val_accuracy: 0.8740
Epoch 51/75
283/283 [==============================] - 14s 47ms/step - loss: 0.2092 - accuracy: 0.9461 - val_loss: 0.5614 - val_accuracy: 0.8753
Epoch 52/75
283/283 [==============================] - 14s 47ms/step - loss: 0.2207 - accuracy: 0.9418 - val_loss: 0.4740 - val_accuracy: 0.8930
Epoch 53/75
283/283 [==============================] - 14s 47ms/step - loss: 0.2077 - accuracy: 0.9423 - val_loss: 1.2104 - val_accuracy: 0.7727
Epoch 54/75
283/283 [==============================] - 14s 48ms/step - loss: 0.2226 - accuracy: 0.9459 - val_loss: 0.4695 - val_accuracy: 0.9160
Epoch 55/75
283/283 [==============================] - 13s 44ms/step - loss: 0.2062 - accuracy: 0.9461 - val_loss: 0.5044 - val_accuracy: 0.8810
Epoch 56/75
283/283 [==============================] - 12s 42ms/step - loss: 0.1865 - accuracy: 0.9474 - val_loss: 0.6067 - val_accuracy: 0.8947
Epoch 57/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2100 - accuracy: 0.9442 - val_loss: 0.7134 - val_accuracy: 0.8640
Epoch 58/75
283/283 [==============================] - 12s 42ms/step - loss: 0.2129 - accuracy: 0.9432 - val_loss: 0.4775 - val_accuracy: 0.8913
94/94 [==============================] - 1s 10ms/step - loss: 0.3305 - accuracy: 0.9190
94/94 [==============================] - 1s 10ms/step - loss: 0.7807 - accuracy: 0.8057
94/94 [==============================] - 1s 10ms/step - loss: 0.2485 - accuracy: 0.9403
94/94 [==============================] - 1s 11ms/step - loss: 0.6039 - accuracy: 0.8300
94/94 [==============================] - 1s 10ms/step - loss: 0.4254 - accuracy: 0.8773
94/94 [==============================] - 1s 10ms/step - loss: 0.4747 - accuracy: 0.8670
94/94 [==============================] - 1s 10ms/step - loss: 2043853066469376.0000 - accuracy: 0.0693
94/94 [==============================] - 1s 10ms/step - loss: 0.9589 - accuracy: 0.7867
94/94 [==============================] - 1s 10ms/step - loss: 0.4775 - accuracy: 0.8913

From the above, we can see that Adam is still the best optimizer for the model, and especially when it has a learning rate of 0.0002. We shall save this model again, so that we do not lose our progress, and run it using the test dataset to see it's final performance. Before we evaluate the model's performance on the test dataset, we shall allow the model to train for another 50 epochs, allow for it to perform to the best of it's ability.

history = tuned_31_Adam_LR_0_0002.fit(aug_res31_train, validation_data = std_res31_validation, epochs = 50, class_weight = class_weights)
cnnf.model_metric_graph(history, tuned_31_Adam_LR_0_0002, std_res128_validation, name = "Extended run for final 128 x 128", class_names = class_names)

Epoch 1/50
283/283 [==============================] - 11s 38ms/step - loss: 0.0639 - accuracy: 0.9809 - val_loss: 0.2376 - val_accuracy: 0.9450
Epoch 2/50
283/283 [==============================] - 11s 38ms/step - loss: 0.0663 - accuracy: 0.9772 - val_loss: 0.2423 - val_accuracy: 0.9427
Epoch 3/50
283/283 [==============================] - 11s 38ms/step - loss: 0.0673 - accuracy: 0.9774 - val_loss: 0.2088 - val_accuracy: 0.9467
Epoch 4/50
283/283 [==============================] - 11s 37ms/step - loss: 0.0540 - accuracy: 0.9822 - val_loss: 0.2266 - val_accuracy: 0.9460
Epoch 5/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0585 - accuracy: 0.9790 - val_loss: 0.2071 - val_accuracy: 0.9500
Epoch 6/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0550 - accuracy: 0.9818 - val_loss: 0.2033 - val_accuracy: 0.9507
Epoch 7/50
283/283 [==============================] - 10s 35ms/step - loss: 0.0519 - accuracy: 0.9822 - val_loss: 0.1963 - val_accuracy: 0.9553
Epoch 8/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0533 - accuracy: 0.9839 - val_loss: 0.2297 - val_accuracy: 0.9463
Epoch 9/50
283/283 [==============================] - 10s 35ms/step - loss: 0.0554 - accuracy: 0.9834 - val_loss: 0.2444 - val_accuracy: 0.9427
Epoch 10/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0520 - accuracy: 0.9842 - val_loss: 0.2245 - val_accuracy: 0.9470
Epoch 11/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0508 - accuracy: 0.9826 - val_loss: 0.2667 - val_accuracy: 0.9373
Epoch 12/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0513 - accuracy: 0.9827 - val_loss: 0.2020 - val_accuracy: 0.9563
Epoch 13/50
283/283 [==============================] - 11s 36ms/step - loss: 0.0507 - accuracy: 0.9822 - val_loss: 0.2094 - val_accuracy: 0.9500
Epoch 14/50
283/283 [==============================] - 10s 35ms/step - loss: 0.0512 - accuracy: 0.9829 - val_loss: 0.2272 - val_accuracy: 0.9477
Epoch 15/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0523 - accuracy: 0.9831 - val_loss: 0.2203 - val_accuracy: 0.9517
Epoch 16/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0542 - accuracy: 0.9821 - val_loss: 0.2630 - val_accuracy: 0.9360
Epoch 17/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0539 - accuracy: 0.9816 - val_loss: 0.2405 - val_accuracy: 0.9443
Epoch 18/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0470 - accuracy: 0.9840 - val_loss: 0.2284 - val_accuracy: 0.9533
Epoch 19/50
283/283 [==============================] - 10s 35ms/step - loss: 0.0561 - accuracy: 0.9821 - val_loss: 0.2367 - val_accuracy: 0.9473
Epoch 20/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0461 - accuracy: 0.9854 - val_loss: 0.3225 - val_accuracy: 0.9257
Epoch 21/50
283/283 [==============================] - 10s 35ms/step - loss: 0.0510 - accuracy: 0.9837 - val_loss: 0.2348 - val_accuracy: 0.9467
Epoch 22/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0630 - accuracy: 0.9798 - val_loss: 0.2272 - val_accuracy: 0.9457
Epoch 23/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0542 - accuracy: 0.9815 - val_loss: 0.2603 - val_accuracy: 0.9417
Epoch 24/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0585 - accuracy: 0.9793 - val_loss: 0.2013 - val_accuracy: 0.9523
Epoch 25/50
283/283 [==============================] - 11s 36ms/step - loss: 0.0472 - accuracy: 0.9831 - val_loss: 0.2163 - val_accuracy: 0.9483
Epoch 26/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0459 - accuracy: 0.9844 - val_loss: 0.2173 - val_accuracy: 0.9527
Epoch 27/50
283/283 [==============================] - 11s 36ms/step - loss: 0.0508 - accuracy: 0.9823 - val_loss: 0.2040 - val_accuracy: 0.9530
Epoch 28/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0538 - accuracy: 0.9834 - val_loss: 0.2195 - val_accuracy: 0.9487
Epoch 29/50
283/283 [==============================] - 11s 36ms/step - loss: 0.0526 - accuracy: 0.9808 - val_loss: 0.2431 - val_accuracy: 0.9433
Epoch 30/50
283/283 [==============================] - 11s 37ms/step - loss: 0.0413 - accuracy: 0.9857 - val_loss: 0.2143 - val_accuracy: 0.9497
Epoch 31/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0509 - accuracy: 0.9814 - val_loss: 0.2019 - val_accuracy: 0.9500
Epoch 32/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0506 - accuracy: 0.9840 - val_loss: 0.1964 - val_accuracy: 0.9520
Epoch 33/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0533 - accuracy: 0.9825 - val_loss: 0.1957 - val_accuracy: 0.9557
Epoch 34/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0414 - accuracy: 0.9856 - val_loss: 0.2283 - val_accuracy: 0.9443
Epoch 35/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0493 - accuracy: 0.9816 - val_loss: 0.2162 - val_accuracy: 0.9477
Epoch 36/50
283/283 [==============================] - 11s 36ms/step - loss: 0.0452 - accuracy: 0.9856 - val_loss: 0.2443 - val_accuracy: 0.9457
Epoch 37/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0498 - accuracy: 0.9825 - val_loss: 0.2377 - val_accuracy: 0.9470
Epoch 38/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0474 - accuracy: 0.9834 - val_loss: 0.2308 - val_accuracy: 0.9470
Epoch 39/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0477 - accuracy: 0.9857 - val_loss: 0.1860 - val_accuracy: 0.9577
Epoch 40/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0547 - accuracy: 0.9844 - val_loss: 0.2185 - val_accuracy: 0.9453
Epoch 41/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0485 - accuracy: 0.9839 - val_loss: 0.2106 - val_accuracy: 0.9510
Epoch 42/50
283/283 [==============================] - 11s 36ms/step - loss: 0.0461 - accuracy: 0.9850 - val_loss: 0.2417 - val_accuracy: 0.9437
Epoch 43/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0387 - accuracy: 0.9875 - val_loss: 0.2355 - val_accuracy: 0.9463
Epoch 44/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0463 - accuracy: 0.9850 - val_loss: 0.2385 - val_accuracy: 0.9490
Epoch 45/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0407 - accuracy: 0.9872 - val_loss: 0.2149 - val_accuracy: 0.9517
Epoch 46/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0456 - accuracy: 0.9847 - val_loss: 0.2272 - val_accuracy: 0.9503
Epoch 47/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0421 - accuracy: 0.9852 - val_loss: 0.2093 - val_accuracy: 0.9527
Epoch 48/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0432 - accuracy: 0.9857 - val_loss: 0.2266 - val_accuracy: 0.9423
Epoch 49/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0451 - accuracy: 0.9846 - val_loss: 0.2115 - val_accuracy: 0.9500
Epoch 50/50
283/283 [==============================] - 10s 36ms/step - loss: 0.0449 - accuracy: 0.9827 - val_loss: 0.2080 - val_accuracy: 0.9507

tuned_31_Adam_LR_0_0002.save("res31_best_tuned.h5")
tuned_31_Adam_LR_0_0002.save_weights("res31_best_tuned_weights.h5")

from sklearn.metrics import accuracy_score
from tensorflow.keras.losses import SparseCategoricalCrossentropy

def predict_in_batches(model, dataset, batch_size=32):
    predictions = []
    true_labels = []

    for batch_data, labels in dataset.as_numpy_iterator():
        batch_predictions = model.predict(batch_data)
        clear_output(wait=True)
        predictions.append(batch_predictions)
        true_labels.append(labels)

    return np.concatenate(predictions, axis=0), np.concatenate(true_labels, axis = 0)

predictions, true_labels = predict_in_batches(tuned_31_Adam_LR_0_0002, std_res31_test, batch_size=32)


scce = SparseCategoricalCrossentropy()
fig = plt.figure(figsize = (20,10))
sns.heatmap(confusion_matrix(true_labels, np.argmax(predictions, axis = 1)), annot = True, xticklabels = class_names, yticklabels = class_names, fmt = "g")
plt.suptitle(f"Final Predictions of model for 31 x 31 images.\n\nAccuracy: {accuracy_score(true_labels, np.argmax(predictions, axis = 1)):.4f}, Loss: {scce(true_labels, predictions).numpy():.4f}")
plt.xlabel('Predicted')
plt.ylabel('True')
plt.show()

We can see that the final model performs quite well, with a accuracy of 94%. We can see that for the most part, the model seems to be able to predict most vegetables correctly, except for Potatoes, which it seems to think are brinjals. Apart from that, the model makes a few mistakes here and there, but is still able to capture most of the detail in the dataset.

---

Model Hypertuning (128 x 128)

---

def model_hypertuner_128(optimizer_list, name_list, model_list):
    for optimizer, name, model in zip(optimizer_list, name_list, model_list):
        clear_output(wait=True)
        print(f"Now attempting to tune {name}")
        model.compile(optimizer = optimizer, loss = "sparse_categorical_crossentropy", metrics=['accuracy'])
        model.fit(aug_res128_train, validation_data = std_res128_validation, epochs = 75, callbacks = earlyStopper, class_weight = class_weights)
        gc.collect()

from tensorflow.keras.optimizers import Adam
from tensorflow.keras.optimizers import SGD
from tensorflow.keras.optimizers import RMSprop
optimizer_list = [
    Adam(),
    Adam(learning_rate=0.005),
    Adam(learning_rate=0.0002),
    SGD(learning_rate=0.005),
    SGD(learning_rate=0.05),
    SGD(),
    RMSprop(learning_rate=0.05),
    RMSprop(learning_rate=0.002),
    RMSprop(),
]

name_list = [
    'Adam_LR_0.001',
    'Adam_LR_0.005',
    'Adam_LR_0.0002',
    'SGD_LR_0.005',
    'SGD_LR_0.05',
    'SGD_LR_0.01',
    'RMSprop_LR_0.05',
    'RMSprop_LR_0.002',
    'RMSprop_LR_0.01'
]

tuned_128_Adam_LR_0_001 = load_model('res128_best.h5')
tuned_128_Adam_LR_0_005 = load_model('res128_best.h5')
tuned_128_Adam_LR_0_0002 = load_model('res128_best.h5')
tuned_128_SGD_LR_0_005 = load_model('res128_best.h5')
tuned_128_SGD_LR_0_05 = load_model('res128_best.h5')
tuned_128_SGD_LR_0_01 = load_model('res128_best.h5')
tuned_128_RMSprop_LR_0_05 = load_model('res128_best.h5')
tuned_128_RMSprop_LR_0_002 = load_model('res128_best.h5')
tuned_128_RMSprop_LR_0_01 = load_model('res128_best.h5')


output_list = [tuned_128_Adam_LR_0_001, tuned_128_Adam_LR_0_005, tuned_128_Adam_LR_0_0002, tuned_128_SGD_LR_0_005, tuned_128_SGD_LR_0_05, tuned_128_SGD_LR_0_01, tuned_128_RMSprop_LR_0_05, tuned_128_RMSprop_LR_0_002, tuned_128_RMSprop_LR_0_01]

data_list = [std_res128_validation, std_res128_validation, std_res128_validation, std_res128_validation, std_res128_validation, std_res128_validation, std_res128_validation, std_res128_validation, std_res128_validation]

model_hypertuner_128(optimizer_list, name_list, output_list)
cnnf.model_comparison(output_list, data_list, name_list)

Now attempting to tune RMSprop_LR_0.01
Epoch 1/75
283/283 [==============================] - 65s 218ms/step - loss: 0.0714 - accuracy: 0.9813 - val_loss: 0.2232 - val_accuracy: 0.9563
Epoch 2/75
283/283 [==============================] - 61s 213ms/step - loss: 0.0524 - accuracy: 0.9854 - val_loss: 0.1875 - val_accuracy: 0.9547
Epoch 3/75
283/283 [==============================] - 62s 217ms/step - loss: 0.0467 - accuracy: 0.9872 - val_loss: 0.1648 - val_accuracy: 0.9663
Epoch 4/75
283/283 [==============================] - 61s 215ms/step - loss: 0.0553 - accuracy: 0.9826 - val_loss: 0.1532 - val_accuracy: 0.9647
Epoch 5/75
283/283 [==============================] - 61s 213ms/step - loss: 0.0464 - accuracy: 0.9868 - val_loss: 0.1049 - val_accuracy: 0.9787
Epoch 6/75
283/283 [==============================] - 63s 220ms/step - loss: 0.0479 - accuracy: 0.9873 - val_loss: 0.1535 - val_accuracy: 0.9723
Epoch 7/75
283/283 [==============================] - 63s 220ms/step - loss: 0.0582 - accuracy: 0.9838 - val_loss: 0.1851 - val_accuracy: 0.9670
Epoch 8/75
283/283 [==============================] - 63s 221ms/step - loss: 0.0494 - accuracy: 0.9875 - val_loss: 0.1479 - val_accuracy: 0.9697
Epoch 9/75
283/283 [==============================] - 62s 218ms/step - loss: 0.0534 - accuracy: 0.9858 - val_loss: 0.4222 - val_accuracy: 0.9210
Epoch 10/75
283/283 [==============================] - 62s 220ms/step - loss: 0.0553 - accuracy: 0.9849 - val_loss: 0.1170 - val_accuracy: 0.9757
Epoch 11/75
283/283 [==============================] - 61s 216ms/step - loss: 0.0586 - accuracy: 0.9858 - val_loss: 0.1213 - val_accuracy: 0.9727
Epoch 12/75
283/283 [==============================] - 62s 217ms/step - loss: 0.0546 - accuracy: 0.9856 - val_loss: 0.1213 - val_accuracy: 0.9700
Epoch 13/75
283/283 [==============================] - 61s 215ms/step - loss: 0.0450 - accuracy: 0.9867 - val_loss: 0.2045 - val_accuracy: 0.9553
Epoch 14/75
283/283 [==============================] - 61s 213ms/step - loss: 0.0563 - accuracy: 0.9850 - val_loss: 0.5783 - val_accuracy: 0.8970
Epoch 15/75
283/283 [==============================] - 62s 218ms/step - loss: 0.0450 - accuracy: 0.9881 - val_loss: 0.1751 - val_accuracy: 0.9643
Epoch 16/75
283/283 [==============================] - 61s 214ms/step - loss: 0.0455 - accuracy: 0.9873 - val_loss: 0.3375 - val_accuracy: 0.9323
Epoch 17/75
283/283 [==============================] - 61s 213ms/step - loss: 0.0534 - accuracy: 0.9866 - val_loss: 0.1415 - val_accuracy: 0.9757
Epoch 18/75
283/283 [==============================] - 62s 219ms/step - loss: 0.0466 - accuracy: 0.9889 - val_loss: 0.2337 - val_accuracy: 0.9607
Epoch 19/75
283/283 [==============================] - 62s 217ms/step - loss: 0.0517 - accuracy: 0.9854 - val_loss: 0.1231 - val_accuracy: 0.9743
Epoch 20/75
283/283 [==============================] - 63s 222ms/step - loss: 0.0504 - accuracy: 0.9875 - val_loss: 0.2232 - val_accuracy: 0.9530
94/94 [==============================] - 5s 45ms/step - loss: 0.0880 - accuracy: 0.9770
94/94 [==============================] - 5s 47ms/step - loss: 0.1283 - accuracy: 0.9677
94/94 [==============================] - 5s 48ms/step - loss: 0.0407 - accuracy: 0.9910
94/94 [==============================] - 5s 47ms/step - loss: 0.0832 - accuracy: 0.9823
94/94 [==============================] - 5s 45ms/step - loss: 0.0718 - accuracy: 0.9823
94/94 [==============================] - 5s 47ms/step - loss: 0.0588 - accuracy: 0.9853
94/94 [==============================] - 5s 48ms/step - loss: 63049.3828 - accuracy: 0.0667
94/94 [==============================] - 5s 47ms/step - loss: 0.8063 - accuracy: 0.8670
94/94 [==============================] - 5s 47ms/step - loss: 0.2232 - accuracy: 0.9530

From the above, we can see that Adam is still the best optimizer for the model, and especially when it has a learning rate of 0.0002. We shall save this model again, so that we do not lose our progress, and run it using the test dataset to see it's final performance. Before we evaluate the model's performance on the test dataset, we shall allow the model to train for another 50 epochs, allow for it to perform to the best of it's ability.

history = tuned_128_Adam_LR_0_0002.fit(aug_res128_train, validation_data = std_res128_validation, epochs = 50, class_weight = class_weights)
cnnf.model_metric_graph(history, tuned_128_Adam_LR_0_0002, std_res128_validation, name = "Extended run for final 128 x 128", class_names = class_names)

Epoch 1/50
283/283 [==============================] - 62s 217ms/step - loss: 0.0115 - accuracy: 0.9967 - val_loss: 0.0422 - val_accuracy: 0.9917
Epoch 2/50
283/283 [==============================] - 59s 208ms/step - loss: 0.0137 - accuracy: 0.9961 - val_loss: 0.0468 - val_accuracy: 0.9897
Epoch 3/50
283/283 [==============================] - 61s 216ms/step - loss: 0.0122 - accuracy: 0.9967 - val_loss: 0.0477 - val_accuracy: 0.9900
Epoch 4/50
283/283 [==============================] - 61s 216ms/step - loss: 0.0094 - accuracy: 0.9977 - val_loss: 0.0389 - val_accuracy: 0.9913
Epoch 5/50
283/283 [==============================] - 62s 216ms/step - loss: 0.0088 - accuracy: 0.9979 - val_loss: 0.0474 - val_accuracy: 0.9893
Epoch 6/50
283/283 [==============================] - 61s 216ms/step - loss: 0.0113 - accuracy: 0.9975 - val_loss: 0.0557 - val_accuracy: 0.9897
Epoch 7/50
283/283 [==============================] - 62s 217ms/step - loss: 0.0117 - accuracy: 0.9969 - val_loss: 0.0529 - val_accuracy: 0.9897
Epoch 8/50
283/283 [==============================] - 62s 217ms/step - loss: 0.0196 - accuracy: 0.9946 - val_loss: 0.0463 - val_accuracy: 0.9890
Epoch 9/50
283/283 [==============================] - 62s 216ms/step - loss: 0.0103 - accuracy: 0.9967 - val_loss: 0.0430 - val_accuracy: 0.9900
Epoch 10/50
283/283 [==============================] - 62s 217ms/step - loss: 0.0111 - accuracy: 0.9972 - val_loss: 0.0455 - val_accuracy: 0.9907
Epoch 11/50
283/283 [==============================] - 62s 216ms/step - loss: 0.0098 - accuracy: 0.9973 - val_loss: 0.0455 - val_accuracy: 0.9897
Epoch 12/50
283/283 [==============================] - 60s 210ms/step - loss: 0.0127 - accuracy: 0.9960 - val_loss: 0.0376 - val_accuracy: 0.9920
Epoch 13/50
283/283 [==============================] - 61s 215ms/step - loss: 0.0133 - accuracy: 0.9966 - val_loss: 0.0340 - val_accuracy: 0.9917
Epoch 14/50
283/283 [==============================] - 62s 217ms/step - loss: 0.0123 - accuracy: 0.9965 - val_loss: 0.0455 - val_accuracy: 0.9907
Epoch 15/50
283/283 [==============================] - 62s 217ms/step - loss: 0.0098 - accuracy: 0.9973 - val_loss: 0.0437 - val_accuracy: 0.9907
Epoch 16/50
283/283 [==============================] - 62s 216ms/step - loss: 0.0092 - accuracy: 0.9972 - val_loss: 0.0388 - val_accuracy: 0.9913
Epoch 17/50
283/283 [==============================] - 62s 218ms/step - loss: 0.0123 - accuracy: 0.9969 - val_loss: 0.0365 - val_accuracy: 0.9907
Epoch 18/50
283/283 [==============================] - 62s 218ms/step - loss: 0.0093 - accuracy: 0.9969 - val_loss: 0.0433 - val_accuracy: 0.9923
Epoch 19/50
283/283 [==============================] - 62s 218ms/step - loss: 0.0117 - accuracy: 0.9966 - val_loss: 0.0376 - val_accuracy: 0.9927
Epoch 20/50
283/283 [==============================] - 62s 218ms/step - loss: 0.0109 - accuracy: 0.9969 - val_loss: 0.0487 - val_accuracy: 0.9900
Epoch 21/50
283/283 [==============================] - 62s 219ms/step - loss: 0.0139 - accuracy: 0.9967 - val_loss: 0.0342 - val_accuracy: 0.9920
Epoch 22/50
283/283 [==============================] - 61s 216ms/step - loss: 0.0130 - accuracy: 0.9962 - val_loss: 0.0337 - val_accuracy: 0.9920
Epoch 23/50
283/283 [==============================] - 61s 215ms/step - loss: 0.0153 - accuracy: 0.9953 - val_loss: 0.0342 - val_accuracy: 0.9917
Epoch 24/50
283/283 [==============================] - 61s 215ms/step - loss: 0.0109 - accuracy: 0.9979 - val_loss: 0.0302 - val_accuracy: 0.9920
Epoch 25/50
283/283 [==============================] - 62s 216ms/step - loss: 0.0092 - accuracy: 0.9971 - val_loss: 0.0372 - val_accuracy: 0.9903
Epoch 26/50
283/283 [==============================] - 62s 218ms/step - loss: 0.0106 - accuracy: 0.9972 - val_loss: 0.0379 - val_accuracy: 0.9903
Epoch 27/50
283/283 [==============================] - 61s 213ms/step - loss: 0.0129 - accuracy: 0.9957 - val_loss: 0.0542 - val_accuracy: 0.9900
Epoch 28/50
283/283 [==============================] - 61s 214ms/step - loss: 0.0081 - accuracy: 0.9979 - val_loss: 0.0480 - val_accuracy: 0.9897
Epoch 29/50
283/283 [==============================] - 61s 213ms/step - loss: 0.0103 - accuracy: 0.9975 - val_loss: 0.0486 - val_accuracy: 0.9907
Epoch 30/50
283/283 [==============================] - 61s 214ms/step - loss: 0.0097 - accuracy: 0.9976 - val_loss: 0.0363 - val_accuracy: 0.9923
Epoch 31/50
283/283 [==============================] - 61s 213ms/step - loss: 0.0115 - accuracy: 0.9965 - val_loss: 0.0469 - val_accuracy: 0.9890
Epoch 32/50
283/283 [==============================] - 61s 216ms/step - loss: 0.0097 - accuracy: 0.9975 - val_loss: 0.0403 - val_accuracy: 0.9907
Epoch 33/50
283/283 [==============================] - 62s 218ms/step - loss: 0.0082 - accuracy: 0.9976 - val_loss: 0.0471 - val_accuracy: 0.9900
Epoch 34/50
283/283 [==============================] - 62s 217ms/step - loss: 0.0111 - accuracy: 0.9967 - val_loss: 0.0397 - val_accuracy: 0.9913
Epoch 35/50
283/283 [==============================] - 62s 216ms/step - loss: 0.0114 - accuracy: 0.9978 - val_loss: 0.0368 - val_accuracy: 0.9907
Epoch 36/50
283/283 [==============================] - 62s 218ms/step - loss: 0.0128 - accuracy: 0.9956 - val_loss: 0.0409 - val_accuracy: 0.9893
Epoch 37/50
283/283 [==============================] - 62s 217ms/step - loss: 0.0073 - accuracy: 0.9984 - val_loss: 0.0397 - val_accuracy: 0.9910
Epoch 38/50
283/283 [==============================] - 62s 217ms/step - loss: 0.0098 - accuracy: 0.9970 - val_loss: 0.0329 - val_accuracy: 0.9923
Epoch 39/50
283/283 [==============================] - 62s 218ms/step - loss: 0.0103 - accuracy: 0.9968 - val_loss: 0.0374 - val_accuracy: 0.9907
Epoch 40/50
283/283 [==============================] - 63s 221ms/step - loss: 0.0093 - accuracy: 0.9977 - val_loss: 0.0400 - val_accuracy: 0.9907
Epoch 41/50
283/283 [==============================] - 65s 227ms/step - loss: 0.0140 - accuracy: 0.9972 - val_loss: 0.0317 - val_accuracy: 0.9923
Epoch 42/50
283/283 [==============================] - 62s 217ms/step - loss: 0.0096 - accuracy: 0.9973 - val_loss: 0.0355 - val_accuracy: 0.9917
Epoch 43/50
283/283 [==============================] - 62s 218ms/step - loss: 0.0069 - accuracy: 0.9981 - val_loss: 0.0387 - val_accuracy: 0.9920
Epoch 44/50
283/283 [==============================] - 62s 218ms/step - loss: 0.0088 - accuracy: 0.9981 - val_loss: 0.0382 - val_accuracy: 0.9907
Epoch 45/50
283/283 [==============================] - 65s 228ms/step - loss: 0.0092 - accuracy: 0.9971 - val_loss: 0.0356 - val_accuracy: 0.9920
Epoch 46/50
283/283 [==============================] - 63s 220ms/step - loss: 0.0083 - accuracy: 0.9972 - val_loss: 0.0385 - val_accuracy: 0.9910
Epoch 47/50
283/283 [==============================] - 63s 223ms/step - loss: 0.0098 - accuracy: 0.9965 - val_loss: 0.0414 - val_accuracy: 0.9887
Epoch 48/50
283/283 [==============================] - 63s 221ms/step - loss: 0.0105 - accuracy: 0.9971 - val_loss: 0.0294 - val_accuracy: 0.9930
Epoch 49/50
283/283 [==============================] - 63s 223ms/step - loss: 0.0091 - accuracy: 0.9973 - val_loss: 0.0310 - val_accuracy: 0.9923
Epoch 50/50
283/283 [==============================] - 63s 222ms/step - loss: 0.0104 - accuracy: 0.9970 - val_loss: 0.0346 - val_accuracy: 0.9913

tuned_128_Adam_LR_0_0002.save("res128_best_tuned.h5")
tuned_128_Adam_LR_0_0002.save_weights("res128_best_tuned_weights.h5")

from sklearn.metrics import accuracy_score
from tensorflow.keras.losses import SparseCategoricalCrossentropy

def predict_in_batches(model, dataset, batch_size=32):
    predictions = []
    true_labels = []

    for batch_data, labels in dataset.as_numpy_iterator():
        batch_predictions = model.predict(batch_data)
        clear_output(wait=True)
        predictions.append(batch_predictions)
        true_labels.append(labels)

    return np.concatenate(predictions, axis=0), np.concatenate(true_labels, axis = 0)

predictions, true_labels = predict_in_batches(tuned_128_Adam_LR_0_0002, std_res128_test, batch_size=32)


scce = SparseCategoricalCrossentropy()
fig = plt.figure(figsize = (20,10))
sns.heatmap(confusion_matrix(true_labels, np.argmax(predictions, axis = 1)), annot = True, xticklabels = class_names, yticklabels = class_names, fmt = "g")
plt.suptitle(f"Final Predictions of model for 128 x 128 images.\n\nAccuracy: {accuracy_score(true_labels, np.argmax(predictions, axis = 1)):.4f}, Loss: {scce(true_labels, predictions).numpy():.4f}")
plt.xlabel('Predicted')
plt.ylabel('True')
plt.show()

The model performs extremely well on the 128 x 128 dataset, capturing most, if not all of the detail in the dataset, and it reflects in the results. With a accuracy of 99.03% on the test dataset, the model is almost perfect. It does make a few mistakes here and there, but that may be due to inconsistencies in the test data.

---

Conclusions

---

To conclude, this exercise has allowed me to learn alot more about the field of deep learning, and the considerations that have to be made when attempting to craft a tool for a task. One part where I understand better are the limitations placed on us, especially regarding the greyscaling and resolution constraints. From the above, it can be seen that the models perform better when given more data (94% on 31 x 31, and 99% on 128 x 128). However, a larger amount of imperfect data is already hard to get, given other constraints, let alone data that is perfectly sized, processed and prepared. From this assignment, we can also see the effects of different ways to improve performance, such as class weights, augmentation etc., and how they affect the model in general.

Saving weights.